usage-tab 1.0.0 → 1.2.0

package/dist/index.cjs

@@ -30,13 +30,15 @@ __export(index_exports, {
   REGISTRY_VERSION: () => REGISTRY_VERSION,
   UnknownModelError: () => UnknownModelError,
   calculateCost: () => calculateCost,
+  createCostAggregator: () => createCostAggregator,
   createPriceCalculator: () => createPriceCalculator,
   createPriceOverride: () => createPriceOverride,
   normalizeAnthropicUsage: () => normalizeAnthropicUsage,
   normalizeGoogleUsage: () => normalizeGoogleUsage,
   normalizeOpenAICompatibleUsage: () => normalizeOpenAICompatibleUsage,
   normalizeOpenAIUsage: () => normalizeOpenAIUsage,
-  resolveModel: () => resolveModel2
+  resolveModel: () => resolveModel2,
+  sumExactUsd: () => sumExactUsd
 });
 module.exports = __toCommonJS(index_exports);
 
@@ -122,8 +124,9 @@ function toCandidate(descriptor) {
 }
 function matchExact(pool, id, provider) {
   if (provider !== void 0) {
-    const canonical = pool.find((d) => d.provider === provider && d.canonicalId === id);
-    if (canonical !== void 0) return { unique: canonical };
+    const canonical = pool.filter((d) => d.provider === provider && d.canonicalId === id);
+    if (canonical.length === 1) return { unique: canonical[0] };
+    if (canonical.length > 1) return { ambiguous: canonical };
     const scoped = pool.filter((d) => d.provider === provider && d.aliases.includes(id));
     if (scoped.length === 1) return { unique: scoped[0] };
     if (scoped.length > 1) return { ambiguous: scoped };
@@ -151,16 +154,22 @@ function resolveModel(requestedId, registry, options = {}) {
     }
   }
   if (provider !== void 0) {
-    const canonical = registry.find(
+    const canonical = registry.filter(
       (d) => d.provider === provider && d.canonicalId === requestedId
     );
-    if (canonical !== void 0) {
-      return {
-        descriptor: canonical,
-        matchedBy: "canonical-qualified",
-        requestedId,
-        requestedProvider: provider
-      };
+    if (canonical.length === 1) {
+      const descriptor = canonical[0];
+      if (descriptor !== void 0) {
+        return {
+          descriptor,
+          matchedBy: "canonical-qualified",
+          requestedId,
+          requestedProvider: provider
+        };
+      }
+    }
+    if (canonical.length > 1) {
+      throw new AmbiguousAliasError(requestedId, canonical.map(toCandidate));
     }
     const scoped = registry.filter(
       (d) => d.provider === provider && d.aliases.includes(requestedId)
@@ -214,10 +223,16 @@ function resolveModel(requestedId, registry, options = {}) {
 }
 
 // ../../internal/model-registry/src/pricing-period.ts
+var OFFSETLESS_ISO_DATETIME = /^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}(:\d{2}(\.\d+)?)?$/;
 function toTimestamp(value) {
-  const ms = value instanceof Date ? value.getTime() : Date.parse(value);
+  if (value instanceof Date) {
+    const ms2 = value.getTime();
+    if (Number.isNaN(ms2)) throw new InvalidLookupDateError(String(value));
+    return ms2;
+  }
+  const ms = Date.parse(OFFSETLESS_ISO_DATETIME.test(value) ? `${value}Z` : value);
   if (Number.isNaN(ms)) {
-    throw new InvalidLookupDateError(value instanceof Date ? value.toISOString() : value);
+    throw new InvalidLookupDateError(value);
   }
   return ms;
 }
@@ -250,7 +265,7 @@ function selectPricingPeriod(periods, at, identity) {
 }
 
 // ../../internal/model-registry/src/generated/registry.ts
-var REGISTRY_VERSION = "registry-5af85ce1a47be918";
+var REGISTRY_VERSION = "registry-e5f4ec7eb681a235";
 var MODEL_REGISTRY = [
   {
     canonicalId: "claude-fable-5",
@@ -259,9 +274,30 @@ var MODEL_REGISTRY = [
     family: "fable",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "1.00", "cacheWrite": "12.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "1.00", "cacheWrite": "12.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-fable-5-1",
+    provider: "anthropic",
+    aliases: [],
+    family: "fable",
+    contextWindow: 1e6,
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "0.25", "cacheWrite": "12.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Cache hits are priced at 0.025x base input on this model (page footnote 1), not the 0.1x every other model uses - cachedInput is read from the table, not derived.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-haiku-3-5",
+    provider: "anthropic",
+    aliases: [],
+    family: "haiku",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.80", "output": "4.00", "cachedInput": "0.08", "cacheWrite": "1.00", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Retired on Anthropic-operated platforms; the pricing page still publishes its rate and notes it remains available on Amazon Bedrock and Google Cloud. Included so historical usage can still be priced.", "This model predates 2026, so the conservative 2026-01-01 effectiveFrom cannot price usage from its actual lifetime; no rollout date was published on the source page. A future pass should source real dates before relying on historical lookups for it.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-haiku-4-5-20251001",
@@ -270,9 +306,59 @@ var MODEL_REGISTRY = [
     family: "haiku",
     contextWindow: 2e5,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.00", "output": "5.00", "cachedInput": "0.10", "cacheWrite": "1.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.00", "output": "5.00", "cachedInput": "0.10", "cacheWrite": "1.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["canonicalId is the dated snapshot id published on the models overview page; claude-haiku-4-5 is the alias that resolves to it."] }
+  },
+  {
+    canonicalId: "claude-mythos-5",
+    provider: "anthropic",
+    aliases: [],
+    family: "mythos",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "1.00", "cacheWrite": "12.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Limited availability model (page links it to anthropic.com/glasswing); its published rate is recorded as-is.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ['canonicalId is the full dated snapshot id; "claude-haiku-4-5" is the short alias Anthropic documents alongside it \u2014 a genuine alias/canonical-ID resolution case.'] }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-mythos-5-1",
+    provider: "anthropic",
+    aliases: [],
+    family: "mythos",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "0.25", "cacheWrite": "12.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Limited availability model (page links it to anthropic.com/glasswing); its published rate is recorded as-is.", "Cache hits are priced at 0.025x base input on this model (page footnote 1), not the 0.1x every other model uses - cachedInput is read from the table, not derived.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-opus-4",
+    provider: "anthropic",
+    aliases: [],
+    family: "opus",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "75.00", "cachedInput": "1.50", "cacheWrite": "18.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Retired on Anthropic-operated platforms; the pricing page still publishes its rate and notes it remains available on Google Cloud. Included so historical usage can still be priced.", "This model predates 2026, so the conservative 2026-01-01 effectiveFrom cannot price usage from its actual lifetime; no rollout date was published on the source page. A future pass should source real dates before relying on historical lookups for it.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-opus-4-1",
+    provider: "anthropic",
+    aliases: [],
+    family: "opus",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "75.00", "cachedInput": "1.50", "cacheWrite": "18.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Retired on Anthropic-operated platforms; the pricing page still publishes its rate and notes it remains available on Amazon Bedrock and Google Cloud. Included so historical usage can still be priced.", "This model predates 2026, so the conservative 2026-01-01 effectiveFrom cannot price usage from its actual lifetime; no rollout date was published on the source page. A future pass should source real dates before relying on historical lookups for it.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-opus-4-5",
+    provider: "anthropic",
+    aliases: [],
+    family: "opus",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-opus-4-6",
@@ -281,9 +367,9 @@ var MODEL_REGISTRY = [
     family: "opus",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-opus-4-7",
@@ -292,9 +378,9 @@ var MODEL_REGISTRY = [
     family: "opus",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-opus-4-8",
@@ -303,9 +389,9 @@ var MODEL_REGISTRY = [
     family: "opus",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-opus-5",
@@ -314,9 +400,29 @@ var MODEL_REGISTRY = [
     family: "opus",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "25.00", "cachedInput": "0.50", "cacheWrite": "6.25", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-sonnet-4",
+    provider: "anthropic",
+    aliases: [],
+    family: "sonnet",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "cacheWrite": "3.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "Retired on Anthropic-operated platforms; the pricing page still publishes its rate and notes it remains available on Amazon Bedrock and Google Cloud. Included so historical usage can still be priced.", "This model predates 2026, so the conservative 2026-01-01 effectiveFrom cannot price usage from its actual lifetime; no rollout date was published on the source page. A future pass should source real dates before relying on historical lookups for it.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "claude-sonnet-4-5",
+    provider: "anthropic",
+    aliases: [],
+    family: "sonnet",
+    pricing: [
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "cacheWrite": "3.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+    ],
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-sonnet-4-6",
@@ -325,9 +431,9 @@ var MODEL_REGISTRY = [
     family: "sonnet",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "cacheWrite": "3.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "cacheWrite": "3.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ["Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05" }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-sonnet-5",
@@ -336,10 +442,9 @@ var MODEL_REGISTRY = [
     family: "sonnet",
     contextWindow: 1e6,
     pricing: [
-      { "effectiveFrom": "2026-01-01", "effectiveTo": "2026-09-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "10.00", "cachedInput": "0.20", "cacheWrite": "2.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Introductory rate, confirmed active through 2026-08-31. This is the golden fixture for effective-date selection (see test/pricing-period.test.ts): a lookup dated 2026-08-15 must select this period.", "effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true introductory-rate start date; only the 2026-08-31 end date was observed.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] },
-      { "effectiveFrom": "2026-09-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "cacheWrite": "3.75", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Standard rate, effective 2026-09-01 immediately after the introductory-rate window (through 2026-08-31) ends. A lookup dated 2026-09-15 must select this period.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "10.00", "cachedInput": "0.20", "cacheWrite": "2.50", "batchMultiplier": "0.5", "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07", "notes": ['$2.00/$10.00 launched as an introductory rate through 2026-08-31. On 2026-09-07 the pricing page states it "is now the standard price" and that "the previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur", so the rate continues open-ended rather than ending 2026-08-31.', "Supersedes the two-period shape recorded on 2026-08-05 (introductory $2.00/$10.00 to 2026-09-01, then standard $3.00/$15.00). That second period was removed, not closed: the higher rate never took effect, so no date range may report it.", "Anthropic did not publish an explicit effective date for this rate as observed; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "cacheWrite reflects the 5-minute cache TTL (1.25x input). The 1-hour TTL write multiplier is 2x input and is not separately modeled by this schema (a single cacheWrite field)."] }
     ],
-    source: { "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "observedAt": "2026-08-05", "notes": ["Two pricing periods on purpose: an introductory rate ($2.00/$10.00) through 2026-08-31, then the standard rate ($3.00/$15.00) from 2026-09-01."] }
+    source: { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "amazon-nova-lite",
@@ -347,7 +452,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "amazon-nova",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "2.40", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Same cross-region/in-region caveat as amazon-nova-micro."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "2.40", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Same cross-region/in-region caveat as amazon-nova-micro.", "Not surfaced by the 2026-09-07 fetch of the same page, whose accessible sections did not include this model. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal."] }
     ],
     source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
   },
@@ -357,7 +462,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "amazon-nova",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "1.20", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": [`AWS's own first-party model (Amazon publishes both Bedrock and Nova), so "aws-bedrock" is effectively first-party pricing here, unlike the resold third-party models in this file.`, "No explicit effective date published for this specific rate (unlike the Claude 3.5 Sonnet rows above, which do carry a stated Dec 2025 date); effectiveFrom is set conservatively to 2026-01-01.", `Bedrock's pricing page also distinguishes "Global cross-region" vs "in-region" inference pricing for Nova; this rate was not confirmed to be specifically the in-region (vs cross-region) figure \u2014 treat as the headline on-demand rate observed.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "1.20", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": [`AWS's own first-party model (Amazon publishes both Bedrock and Nova), so "aws-bedrock" is effectively first-party pricing here, unlike the resold third-party models in this file.`, "No explicit effective date published for this specific rate (unlike the Claude 3.5 Sonnet rows above, which do carry a stated Dec 2025 date); effectiveFrom is set conservatively to 2026-01-01.", `Bedrock's pricing page also distinguishes "Global cross-region" vs "in-region" inference pricing for Nova; this rate was not confirmed to be specifically the in-region (vs cross-region) figure \u2014 treat as the headline on-demand rate observed.`, "Not surfaced by the 2026-09-07 fetch of the same page, whose accessible sections did not include this model. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal."] }
     ],
     source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
   },
@@ -367,7 +472,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "amazon-nova",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.20", "output": "4.80", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Same cross-region/in-region caveat as amazon-nova-micro."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.20", "output": "4.80", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Same cross-region/in-region caveat as amazon-nova-micro.", "Not surfaced by the 2026-09-07 fetch of the same page, whose accessible sections did not include this model. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal."] }
     ],
     source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
   },
@@ -377,9 +482,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "anthropic-claude",
     pricing: [
-      { "effectiveFrom": "2025-12-01", "currency": "USD", "unit": "per-million-tokens", "input": "6.00", "output": "30.00", "batchMultiplier": "0.5", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": [`AWS Bedrock's own rate card for this Anthropic model, explicitly labeled "Effective 1 Dec 2025" on the pricing page \u2014 a real, confirmed effective date, not the conservative default used elsewhere in this file.`, "Batch: $3.00/$15.00 (confirmed 0.5x standard) as observed on the same page.", "This is Bedrock's own price for an older Claude generation (3.5 Sonnet), not the current claude-sonnet-5 model in anthropic.json \u2014 no comparable first-party entry exists in this registry for the same model, so no direct parity claim is possible or intended. Bedrock and Azure resell other vendors' models under their own rate cards, so a first-party price is never a safe proxy for theirs."] }
+      { "effectiveFrom": "2025-12-01", "currency": "USD", "unit": "per-million-tokens", "input": "6.00", "output": "30.00", "batchMultiplier": "0.5", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": [`AWS Bedrock's own rate card for this Anthropic model, explicitly labeled "Effective 1 Dec 2025" on the pricing page \u2014 a real, confirmed effective date, not the conservative default used elsewhere in this file.`, "Batch: $3.00/$15.00 (confirmed 0.5x standard) as observed on the same page.", "This is Bedrock's own price for an older Claude generation (3.5 Sonnet), not the current claude-sonnet-5 model in anthropic.json \u2014 no comparable first-party entry exists in this registry for the same model, so no direct parity claim is possible or intended. Bedrock and Azure resell other vendors' models under their own rate cards, so a first-party price is never a safe proxy for theirs.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "claude-3.5-sonnet-v2",
@@ -387,9 +492,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "anthropic-claude",
     pricing: [
-      { "effectiveFrom": "2025-12-01", "currency": "USD", "unit": "per-million-tokens", "input": "6.00", "output": "30.00", "cachedInput": "0.60", "cacheWrite": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": [`AWS Bedrock's own rate card, explicitly labeled "Effective 1 Dec 2025" on the pricing page.`, "Batch: $3.00/$15.00 (confirmed 0.5x standard) as observed on the same page.", "Bedrock's own price for an older Claude generation; not comparable to any first-party entry currently in anthropic.json (which covers 4.x/5.x models only)."] }
+      { "effectiveFrom": "2025-12-01", "currency": "USD", "unit": "per-million-tokens", "input": "6.00", "output": "30.00", "cachedInput": "0.60", "cacheWrite": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": [`AWS Bedrock's own rate card, explicitly labeled "Effective 1 Dec 2025" on the pricing page.`, "Batch: $3.00/$15.00 (confirmed 0.5x standard) as observed on the same page.", "Bedrock's own price for an older Claude generation; not comparable to any first-party entry currently in anthropic.json (which covers 4.x/5.x models only).", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "gemma-3-12b",
+    provider: "aws-bedrock",
+    aliases: [],
+    family: "gemma",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.09", "output": "0.29", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": ["Bedrock's own resale rate for this google model, read from Bedrock's pricing page - never copied from that vendor's first-party file.", "Added from the 2026-09-07 fetch; effectiveFrom is the observation date, since the 2026-08-05 fetch of the same page did not surface it.", "US East on-demand rate as printed; Bedrock prices some models per region and this schema has no region dimension."] }
+    ],
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "gemma-3-27b",
+    provider: "aws-bedrock",
+    aliases: [],
+    family: "gemma",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.23", "output": "0.38", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": ["Bedrock's own resale rate for this google model, read from Bedrock's pricing page - never copied from that vendor's first-party file.", "Added from the 2026-09-07 fetch; effectiveFrom is the observation date, since the 2026-08-05 fetch of the same page did not surface it.", "US East on-demand rate as printed; Bedrock prices some models per region and this schema has no region dimension."] }
+    ],
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemma-4-31b",
@@ -397,9 +522,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "google-gemma",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.14", "output": "0.40", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Together AI also lists "Gemma 4 31B" (together.json: gemma-4-31b) at a different, higher rate ($0.39/$0.97 as observed) \u2014 same underlying Google open-weight model, independently priced by each reseller; do not assume parity.'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.14", "output": "0.40", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Together AI also lists "Gemma 4 31B" (together.json: gemma-4-31b) at a different, higher rate ($0.39/$0.97 as observed) \u2014 same underlying Google open-weight model, independently priced by each reseller; do not assume parity.', "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "mistral-large-3",
@@ -407,9 +532,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "mistral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Cross-provider alias/canonicalId collision (allowed, not an error): Mistral's own first-party pricing (mistral.json: mistral-large-3) shows the identical $0.50/$1.50 figure as independently observed \u2014 coincidental agreement between the two independently fetched sources, not assumed; both were confirmed directly."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Cross-provider alias/canonicalId collision (allowed, not an error): Mistral's own first-party pricing (mistral.json: mistral-large-3) shows the identical $0.50/$1.50 figure as independently observed \u2014 coincidental agreement between the two independently fetched sources, not assumed; both were confirmed directly.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "nemotron-3-super-120b",
+    provider: "aws-bedrock",
+    aliases: [],
+    family: "nemotron",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.65", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07", "notes": ["Bedrock's own resale rate for this nvidia model, read from Bedrock's pricing page - never copied from that vendor's first-party file.", "Added from the 2026-09-07 fetch; effectiveFrom is the observation date, since the 2026-08-05 fetch of the same page did not surface it.", "US East on-demand rate as printed; Bedrock prices some models per region and this schema has no region dimension."] }
+    ],
+    source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "nemotron-nano-2",
@@ -417,7 +552,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "nvidia-nemotron",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.06", "output": "0.23", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.06", "output": "0.23", "sourceUrl": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05", "notes": ["No explicit effective date published for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Not surfaced by the 2026-09-07 fetch of the same page, whose accessible sections did not include this model. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal."] }
     ],
     source: { "url": "https://aws.amazon.com/bedrock/pricing/", "observedAt": "2026-08-05" }
   },
@@ -647,9 +782,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.6",
     pricing: [
-      { "effectiveFrom": "2026-07-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "cacheWrite": "6.25", "cheapestTier": true, "sourceUrl": "https://prices.azure.com/api/retail/prices?currencyCode='USD'&$filter=contains(productName,%20%27OpenAI%27)", "observedAt": "2026-08-05", "notes": ["retailPrice confirmed identical across all 24 Azure regions returned for this Global-deployment SKU (spot-checked programmatically, not just a couple of regions) \u2014 no regional price variation observed for the Global tier.", "No Batch API meter was found for this model/SKU in the Retail Prices API response; batchMultiplier is intentionally omitted rather than assumed.", `Matches OpenAI's own first-party rate for "gpt-5.6-sol" in openai.json exactly as observed ($5.00/$30.00/$0.50 cached) \u2014 no markup detected for this model on Azure's Global deployment tier.`, `canonicalId "gpt-5.6-sol" is deliberately identical to the same model's entry in openai.json \u2014 Azure genuinely resells the identical first-party OpenAI model, unlike AWS Bedrock's or OpenRouter's differently-branded catalogues. Following the aws-bedrock.json precedent (e.g. its "mistral-large-3" and "gemma-4-31b" entries) rather than OpenRouter's provider-prefixed-slug convention: this is an intentional cross-provider canonicalId collision, not an error. A bare lookup of this id without a provider qualifier is ambiguous by design and must fail, exactly as already documented for the aws-bedrock.json collisions, since a lookup with a duplicated canonicalId requires a provider qualifier to resolve unambiguously. Per this task's ownership boundary, openai.json itself was not modified to cross-reference this note; a follow-up pass should add a mirroring note there.`] }
+      { "effectiveFrom": "2026-07-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "cacheWrite": "6.25", "cheapestTier": true, "sourceUrl": "https://prices.azure.com/api/retail/prices?currencyCode='USD'&$filter=contains(productName,%20%27OpenAI%27)", "observedAt": "2026-09-07", "notes": ["retailPrice confirmed identical across all 24 Azure regions returned for this Global-deployment SKU (spot-checked programmatically, not just a couple of regions) \u2014 no regional price variation observed for the Global tier.", "No Batch API meter was found for this model/SKU in the Retail Prices API response; batchMultiplier is intentionally omitted rather than assumed.", `canonicalId "gpt-5.6-sol" is deliberately identical to the same model's entry in openai.json \u2014 Azure genuinely resells the identical first-party OpenAI model, unlike AWS Bedrock's or OpenRouter's differently-branded catalogues. Following the aws-bedrock.json precedent (e.g. its "mistral-large-3" and "gemma-4-31b" entries) rather than OpenRouter's provider-prefixed-slug convention: this is an intentional cross-provider canonicalId collision, not an error. A bare lookup of this id without a provider qualifier is ambiguous by design and must fail, exactly as already documented for the aws-bedrock.json collisions, since a lookup with a duplicated canonicalId requires a provider qualifier to resolve unambiguously. Per this task's ownership boundary, openai.json itself was not modified to cross-reference this note; a follow-up pass should add a mirroring note there.`, `Re-observed 2026-09-07 via the Retail Prices API filtered to this model's meters ("5.6 sol ShortCo Inp Std Gl" $5.00, "5.6 sol ShortCo Opt Std Gl" $30.00, "5.6 sol ShortCo Cd Inp Std Gl" $0.50): unchanged.`, `NOW DIFFERS from OpenAI's own first-party rate in openai.json for the same canonicalId "gpt-5.6-sol". Both files recorded $5.00/$30.00/$0.50 on 2026-08-05; on 2026-09-07 OpenAI's pricing page published $4.00/$20.00/$0.40 while Azure's meters stayed at $5.00/$30.00/$0.50. Azure did not follow the first-party cut, so this joins gpt-5.6-terra and gpt-5.6-luna as a confirmed same-id price divergence rather than a transcription error.`, `The API also exposes "LongCo" (long-context) meters for this model at $10.00 input / $45.00 output Global Standard, alongside the "ShortCo" rates recorded here - the same context tiering OpenAI's page labels "<272K". This schema has no context-length dimension, so only the ShortCo tier is recorded.`] }
     ],
-    source: { "url": "https://prices.azure.com/api/retail/prices?currencyCode='USD'&$filter=contains(productName,%20%27OpenAI%27)", "observedAt": "2026-08-05" }
+    source: { "url": "https://prices.azure.com/api/retail/prices?currencyCode='USD'&$filter=contains(productName,%20%27OpenAI%27)", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.6-terra",
@@ -747,9 +882,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "aya-expanse",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": ["Cohere's pricing page publishes an identical $0.50/$1.50 rate for both the 8B and 32B Aya Expanse sizes (confirmed by two independent re-fetches of the same page); this was double-checked rather than assumed to be an extraction error.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": ["Cohere's pricing page publishes an identical $0.50/$1.50 rate for both the 8B and 32B Aya Expanse sizes (confirmed by two independent re-fetches of the same page); this was double-checked rather than assumed to be an extraction error.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "aya-expanse-8b",
@@ -757,9 +892,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "aya-expanse",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": ["Cohere's pricing page publishes an identical $0.50/$1.50 rate for both the 8B and 32B Aya Expanse sizes (confirmed by two independent re-fetches of the same page); this was double-checked rather than assumed to be an extraction error.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": ["Cohere's pricing page publishes an identical $0.50/$1.50 rate for both the 8B and 32B Aya Expanse sizes (confirmed by two independent re-fetches of the same page); this was double-checked rather than assumed to be an extraction error.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "command",
@@ -767,9 +902,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "command",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.00", "output": "2.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": ["Cohere's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "This rate appears in the pricing page's FAQ/legacy-rates section, not a headline pricing table; it is nonetheless the only per-token price Cohere currently publishes for this model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.00", "output": "2.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": ["Cohere's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "This rate appears in the pricing page's FAQ/legacy-rates section, not a headline pricing table; it is nonetheless the only per-token price Cohere currently publishes for this model.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "command-light",
@@ -777,9 +912,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "command",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "0.60", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": ["Cohere's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'FAQ/legacy-rates section pricing, per the same caveat as "command".'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "0.60", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": ["Cohere's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'FAQ/legacy-rates section pricing, per the same caveat as "command".', "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "command-r-03-2024",
@@ -787,9 +922,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "command-r",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": [`"03-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01 per this repository's convention (a too-early effectiveFrom is safe; the price could have applied earlier than 2026 but was not independently confirmed).`, "FAQ/legacy-rates section pricing."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": [`"03-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01 per this repository's convention (a too-early effectiveFrom is safe; the price could have applied earlier than 2026 but was not independently confirmed).`, "FAQ/legacy-rates section pricing.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "command-r-plus-04-2024",
@@ -797,9 +932,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "command-r-plus",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": [`"04-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01.`, `FAQ/legacy-rates section pricing. Superseded in Cohere's catalogue by "Command R+ 08-2024" (below), a distinct dated snapshot with its own price \u2014 the two are not the same PricingPeriod for one model, they are two different canonicalIds, matching how Cohere itself lists them.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": [`"04-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01.`, `FAQ/legacy-rates section pricing. Superseded in Cohere's catalogue by "Command R+ 08-2024" (below), a distinct dated snapshot with its own price \u2014 the two are not the same PricingPeriod for one model, they are two different canonicalIds, matching how Cohere itself lists them.`, "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "command-r-plus-08-2024",
@@ -807,9 +942,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "command-r-plus",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "10.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-08-05", "notes": [`"08-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01.`, "FAQ/legacy-rates section pricing."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "10.00", "sourceUrl": "https://cohere.com/pricing", "observedAt": "2026-09-07", "notes": [`"08-2024" is Cohere's own dated-snapshot naming for this model, not a confirmed price-effective date; effectiveFrom is set conservatively to 2026-01-01.`, "FAQ/legacy-rates section pricing.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://cohere.com/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-2.5-flash",
@@ -817,9 +952,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-2.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "2.50", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Standard (paid) tier rate. Priority tier is $0.54/$4.50 (1.8x standard); not modeled as a separate field.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published.", "cachedInput omitted: not confirmed per-model.", "batchMultiplier of 0.5 verified directly from this model's own Batch row ($0.15/$1.25 vs standard $0.30/$2.50)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "2.50", "cachedInput": "0.03", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "Input and context-caching rates are the text/image/video rates. The page prices audio input separately ($1.00 input, $0.10 context caching); this schema has no per-modality dimension, so the audio rate is not recorded.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.15 / $1.25), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-2.5-flash-lite",
@@ -827,9 +962,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-2.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.40", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Standard (paid) tier rate.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published.", "cachedInput omitted: not confirmed per-model.", "batchMultiplier of 0.5 verified directly from this model's own Batch row ($0.05/$0.20 vs standard $0.10/$0.40)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.40", "cachedInput": "0.01", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "Input and context-caching rates are the text/image/video rates. The page prices audio input separately ($0.30 input, $0.03 context caching); this schema has no per-modality dimension, so the audio rate is not recorded.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.05 / $0.20), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-2.5-pro",
@@ -837,9 +972,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-2.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cheapestTier": true, "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["This is the standard-tier rate for prompts <= 200k tokens. For prompts > 200k tokens the page publishes a higher rate ($2.50 input / $15.00 output per 1M tokens) \u2014 this schema has no context-length-tiered pricing field, so only the <=200k (lower) tier is recorded here. Do not use this entry for long-context (>200k) requests.", "cachedInput and batchMultiplier are omitted: not confirmed for this Pro-tier model (the page states Batch/Flex give a general 50% reduction on input/output pricing, but no explicit per-model Batch row for this model was independently verified).", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "batchMultiplier": "0.5", "cheapestTier": true, "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "The page prices prompts >200k tokens higher ($2.50 input / $15.00 output / $0.25 context caching); only the <=200k tier is recorded, since this schema has no context-length dimension.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.625 / $5.00), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model.", "batchMultiplier and cachedInput added on 2026-09-07: the page now publishes this model's own Batch and context-caching rows, which were not confirmed per-model on 2026-08-05."] }
+    ],
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "gemini-3.1-flash-lite",
+    provider: "google",
+    aliases: [],
+    family: "gemini-3.1",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.25", "output": "1.50", "cachedInput": "0.025", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["New model: absent from the pricing page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than a backdated guess.", "Input and context-caching rates are the text/image/video rates. The page prices audio input separately ($0.50 input, $0.05 context caching); this schema has no per-modality dimension, so the audio rate is not recorded.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.125 / $0.75), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-3.1-pro-preview",
@@ -847,9 +992,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-3.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cheapestTier": true, "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["This is the standard-tier rate for prompts <= 200k tokens. For prompts > 200k tokens the page publishes a higher rate ($4.00 input / $18.00 output per 1M tokens) \u2014 this schema has no context-length-tiered pricing field, so only the <=200k (lower) tier is recorded here. Do not use this entry for long-context (>200k) requests.", "cachedInput and batchMultiplier are omitted: not confirmed for this Pro-tier model (unlike the Flash-tier models above, no explicit per-model Batch row was found for this model).", 'effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published. "-preview" in the model name suggests this may be short-lived/subject to change.', 'canonicalId uses the exact model name Google publishes on the pricing page ("Gemini 3.1 Pro Preview").'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cachedInput": "0.20", "batchMultiplier": "0.5", "cheapestTier": true, "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "The page prices prompts >200k tokens higher ($4.00 input / $18.00 output / $0.40 context caching); only the <=200k tier is recorded, since this schema has no context-length dimension.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($1.00 / $6.00), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model.", "batchMultiplier and cachedInput added on 2026-09-07: the page now publishes this model's own Batch and context-caching rows, which were not confirmed per-model on 2026-08-05."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-3.5-flash",
@@ -857,9 +1002,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-3.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "9.00", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Standard (paid) tier rate. Priority tier is $2.70/$16.20 (1.8x standard); not modeled as a separate field.", 'effectiveFrom set conservatively to 2026-01-01; page shows a "Last Updated: July 30, 2026" stamp but not a rate-specific effective date.', "cachedInput omitted: not confirmed per-model (see gemini-3.6-flash notes for the same caveat).", "batchMultiplier of 0.5 verified directly from this model's own Batch row ($0.75/$4.50 vs standard $1.50/$9.00)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "9.00", "cachedInput": "0.15", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.75 / $4.50), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model.", "cachedInput added on 2026-09-07: the page now publishes a per-model context-caching rate, which it did not on 2026-08-05."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-3.5-flash-lite",
@@ -867,9 +1012,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-3.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "2.50", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Standard (paid) tier rate. Priority tier is $0.54/$4.50 (1.8x standard); not modeled as a separate field.", "effectiveFrom set conservatively to 2026-01-01; exact rate-effective date not published.", "cachedInput omitted: not confirmed per-model.", "batchMultiplier of 0.5 verified directly from this model's own Batch row ($0.15/$1.25 vs standard $0.30/$2.50)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "2.50", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.15 / $1.25), not from a blanket statement.", "No context-caching rate is published for this model; cachedInput is omitted rather than inferred from a sibling model."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gemini-3.6-flash",
@@ -877,9 +1022,33 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gemini-3.6",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Standard (paid) tier rate. The page also lists Flex and Priority tiers, which this schema does not model as separate fields: Flex is priced the same as Batch ($0.75/$3.75); Priority is $2.70/$13.50 (1.8x standard).", `Google's page shows "Last Updated: July 30, 2026 UTC" but does not state when this specific rate took effect; effectiveFrom is set conservatively to 2026-01-01.`, `cachedInput (context caching) is omitted: the page states a general "$0.15 per 1M cached input tokens" figure covering multiple models but does not confirm it is this specific model's rate, plus a separate per-hour storage fee this schema does not model. Recording an unconfirmed number would be worse than omitting it.`, "batchMultiplier of 0.5 was verified directly from this model's own Batch row ($0.75/$3.75 vs standard $1.50/$7.50)."] }
+      { "effectiveFrom": "2026-01-01", "effectiveTo": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Supersedes the single $1.50 / $7.50 period recorded on 2026-08-05. That rate was live then and is scheduled to return on 2027-01-01, so it is kept as a closed historical period rather than deleted.", "Recorded 2026-08-05 at $1.50 / $7.50 with no cachedInput; the page did not then publish a per-model context-caching rate. Closed at the 2026-09-07 observation date, the last date the promotional rate is known not to have applied being 2026-08-05.", "Google's page does not state when this rate took effect; effectiveFrom is set conservatively to 2026-01-01."] },
+      { "effectiveFrom": "2026-09-07", "effectiveTo": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.75", "output": "3.75", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Promotional rate published as running through 2026-12-31, with the standard rate resuming 2027-01-01 - both dates are stated on the page, so this period's effectiveTo and the next period's effectiveFrom are sourced, not conservative placeholders.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] },
+      { "effectiveFrom": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "cachedInput": "0.15", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Standard rate resuming 2027-01-01, the date the page publishes for the end of the promotional rate.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
     ],
-    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Three periods: the $1.50/$7.50 rate observed 2026-08-05, the $0.75/$3.75 promotional rate observed 2026-09-07 and published as running through 2026-12-31, then the standard rate resuming 2027-01-01."] }
+  },
+  {
+    canonicalId: "gemini-3.7-flash",
+    provider: "google",
+    aliases: [],
+    family: "gemini-3.7",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "effectiveTo": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.75", "output": "3.75", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["New model: absent from the pricing page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than a backdated guess.", "Promotional rate published as running through 2026-12-31, with the standard rate resuming 2027-01-01 - both dates are stated on the page, so this period's effectiveTo and the next period's effectiveFrom are sourced, not conservative placeholders.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] },
+      { "effectiveFrom": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "cachedInput": "0.15", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Standard rate resuming 2027-01-01, the date the page publishes for the end of the promotional rate.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
+    ],
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "gemini-3.8-flash",
+    provider: "google",
+    aliases: [],
+    family: "gemini-3.8",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "effectiveTo": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.75", "output": "3.75", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["New model: absent from the pricing page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than a backdated guess.", "Promotional rate published as running through 2026-12-31, with the standard rate resuming 2027-01-01 - both dates are stated on the page, so this period's effectiveTo and the next period's effectiveFrom are sourced, not conservative placeholders.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] },
+      { "effectiveFrom": "2027-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "cachedInput": "0.15", "batchMultiplier": "0.5", "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Standard rate resuming 2027-01-01, the date the page publishes for the end of the promotional rate.", "batchMultiplier of 0.5 was verified from this model's own Batch rows ($0.375 / $1.875 promo, $0.75 / $3.75 standard), not from a blanket statement.", "cachedInput is this model's own published context-caching rate. The page also charges a separate per-hour cache storage fee, which this per-token schema does not model."] }
+    ],
+    source: { "url": "https://ai.google.dev/gemini-api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-oss-120b",
@@ -887,9 +1056,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-oss",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page.", "OpenAI's open-weight gpt-oss-120b model, hosted independently by Groq under Groq's own rate card (also hosted by Together AI, at the same $0.15/$0.60 rate as observed \u2014 coincidental agreement, not assumed parity)."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-09-07", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page.", "OpenAI's open-weight gpt-oss-120b model, hosted independently by Groq under Groq's own rate card (also hosted by Together AI, at the same $0.15/$0.60 rate as observed \u2014 coincidental agreement, not assumed parity).", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-08-05" }
+    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-oss-20b",
@@ -897,9 +1066,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-oss",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.075", "output": "0.30", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page.", "Also hosted by Together AI at a different rate ($0.05/$0.20 as observed) \u2014 each host prices it independently; do not assume parity."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.075", "output": "0.30", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-09-07", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page.", "Also hosted by Together AI at a different rate ($0.05/$0.20 as observed) \u2014 each host prices it independently; do not assume parity.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-08-05" }
+    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "llama-3.1-8b-instant",
@@ -907,7 +1076,7 @@ var MODEL_REGISTRY = [
     aliases: ["llama-3.1-8b"],
     family: "llama-3.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.08", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page, so cachedInput and batchMultiplier are omitted rather than assumed."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.08", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page, so cachedInput and batchMultiplier are omitted rather than assumed.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the gpt-oss and Qwen models with per-token prices. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-08-05" }
   },
@@ -917,7 +1086,7 @@ var MODEL_REGISTRY = [
     aliases: ["llama-3.3-70b"],
     family: "llama-3.3",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.59", "output": "0.79", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page, so cachedInput and batchMultiplier are omitted rather than assumed.", "This is Groq's own hosted rate for the same open-weight model Together AI also hosts (see together.json's llama-3.3-70b at a different, higher price) and AWS Bedrock resells \u2014 each host prices it independently; do not assume parity."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.59", "output": "0.79", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": ["Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No prompt-caching or Batch API discount is documented for Groq in the fetched page, so cachedInput and batchMultiplier are omitted rather than assumed.", "This is Groq's own hosted rate for the same open-weight model Together AI also hosts (see together.json's llama-3.3-70b at a different, higher price) and AWS Bedrock resells \u2014 each host prices it independently; do not assume parity.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the gpt-oss and Qwen models with per-token prices. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-08-05" }
   },
@@ -927,19 +1096,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "qwen",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "3.00", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-08-05", "notes": [`Listed under Groq's "Preview" models section, not "Production" \u2014 preview models on Groq are explicitly subject to change or removal without notice. Included because a real price was published, but treat this one as less stable than the production-tier entries in this file.`, "Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "3.00", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-09-07", "notes": [`Listed under Groq's "Preview" models section, not "Production" \u2014 preview models on Groq are explicitly subject to change or removal without notice. Included because a real price was published, but treat this one as less stable than the production-tier entries in this file.`, "Groq's docs page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-08-05" }
+    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.8-27b",
+    provider: "groq",
+    aliases: ["qwen/qwen3.8-27b"],
+    family: "qwen3.8",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.80", "output": "4.00", "sourceUrl": "https://console.groq.com/docs/models", "observedAt": "2026-09-07", "notes": ["New model: absent from this page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the earlier fetch proves it was not listed then.", "Marked Preview on Groq's page - less stable than a Production model, and its price may move accordingly.", "No cached-input rate and no batch discount are published for Groq models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://console.groq.com/docs/models", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "codestral",
     provider: "mistral",
-    aliases: [],
+    aliases: ["codestral-latest"],
     family: "codestral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "0.90", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "0.90", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'codestral-latest' (recorded as an alias)."] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "devstral-2",
@@ -947,7 +1126,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "devstral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.40", "output": "2.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.40", "output": "2.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
@@ -957,7 +1136,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "devstral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.30", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.30", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
@@ -967,7 +1146,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "magistral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "5.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Listed as a reasoning model on the pricing page; no separate "reasoning" surcharge is published, so the reasoning field is omitted.'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "5.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Listed as a reasoning model on the pricing page; no separate "reasoning" surcharge is published, so the reasoning field is omitted.', "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
@@ -977,59 +1156,59 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "magistral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
   {
     canonicalId: "ministral-3-14b",
     provider: "mistral",
-    aliases: [],
+    aliases: ["ministral-14b-latest"],
     family: "ministral-3",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "0.20", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", `AWS Bedrock resells a "Ministral 14B 3.0" at the same $0.20/$0.20 figure as independently observed on Bedrock's pricing page \u2014 likely the same model, coincidental agreement not assumed; Bedrock's variant was not added to aws-bedrock.json in this pass since the version suffix ("3.0") was not cross-checked against this "ministral-3-14b" naming.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "0.20", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", `AWS Bedrock resells a "Ministral 14B 3.0" at the same $0.20/$0.20 figure as independently observed on Bedrock's pricing page \u2014 likely the same model, coincidental agreement not assumed; Bedrock's variant was not added to aws-bedrock.json in this pass since the version suffix ("3.0") was not cross-checked against this "ministral-3-14b" naming.`, "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'ministral-14b-latest' (recorded as an alias)."] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "ministral-3-3b",
     provider: "mistral",
-    aliases: [],
+    aliases: ["ministral-3b-latest"],
     family: "ministral-3",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.10", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.10", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'ministral-3b-latest' (recorded as an alias)."] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "ministral-3-8b",
     provider: "mistral",
-    aliases: [],
+    aliases: ["ministral-8b-latest"],
     family: "ministral-3",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.15", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.15", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'ministral-8b-latest' (recorded as an alias)."] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "mistral-large-3",
     provider: "mistral",
-    aliases: [],
+    aliases: ["mistral-large-latest"],
     family: "mistral-large",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", `This is Mistral's own first-party rate. AWS Bedrock also resells "Mistral Large 3" under its own rate card at the same $0.50/$1.50 figure as independently observed on Bedrock's pricing page \u2014 coincidental agreement between the two sources, not assumed; see aws-bedrock.json.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "cachedInput": "0.05", "batchMultiplier": "0.5", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", `This is Mistral's own first-party rate. AWS Bedrock also resells "Mistral Large 3" under its own rate card at the same $0.50/$1.50 figure as independently observed on Bedrock's pricing page \u2014 coincidental agreement between the two sources, not assumed; see aws-bedrock.json.`, "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'mistral-large-latest' (recorded as an alias).", `cachedInput and batchMultiplier added on 2026-09-07: the page publishes "Cached input: 90% discount" and "Batch: 50% discount" for this model, so cachedInput is 0.1x this model's own input rate (0.50 -> 0.05), computed as an exact decimal, and batchMultiplier is 0.5. Both come from this model's own row, not from a sibling's rate.`] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "mistral-medium-3.5",
     provider: "mistral",
-    aliases: [],
+    aliases: ["mistral-medium-latest"],
     family: "mistral-medium",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No cachedInput or batchMultiplier is documented on the fetched page for this model; omitted rather than assumed."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.50", "output": "7.50", "cachedInput": "0.15", "batchMultiplier": "0.5", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "No cachedInput or batchMultiplier is documented on the fetched page for this model; omitted rather than assumed.", "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'mistral-medium-latest' (recorded as an alias).", `cachedInput and batchMultiplier added on 2026-09-07: the page publishes "Cached input: 90% discount" and "Batch: 50% discount" for this model, so cachedInput is 0.1x this model's own input rate (1.50 -> 0.15), computed as an exact decimal, and batchMultiplier is 0.5. Both come from this model's own row, not from a sibling's rate.`] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "mistral-nemo",
@@ -1037,19 +1216,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "mistral-nemo",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.15", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched; included because it is confidently sourced, not because it is a current flagship."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.15", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched; included because it is confidently sourced, not because it is a current flagship.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
   {
     canonicalId: "mistral-small-4",
     provider: "mistral",
-    aliases: [],
+    aliases: ["mistral-small-latest"],
     family: "mistral-small",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "cachedInput": "0.015", "batchMultiplier": "0.5", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page, which prints the API id as 'mistral-small-latest' (recorded as an alias).", `cachedInput and batchMultiplier added on 2026-09-07: the page publishes "Cached input: 90% discount" and "Batch: 50% discount" for this model, so cachedInput is 0.1x this model's own input rate (0.15 -> 0.015), computed as an exact decimal, and batchMultiplier is 0.5. Both come from this model's own row, not from a sibling's rate.`] }
     ],
-    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "mixtral-8x22b",
@@ -1057,7 +1236,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "mixtral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "6.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "6.00", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
@@ -1067,19 +1246,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "mixtral",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.70", "output": "0.70", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.70", "output": "0.70", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05", "notes": ["Mistral's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "An older model generation still listed with a live price on the current pricing page as fetched.", "Not surfaced by the 2026-09-07 fetch of the same page, which listed only the Medium/Small/Large, Ministral, Codestral, embedding and third-party GLM entries. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal. Re-check before relying on this entry."] }
     ],
     source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-08-05" }
   },
+  {
+    canonicalId: "zai-glm-5-2",
+    provider: "mistral",
+    aliases: [],
+    family: "glm",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "1.40", "output": "4.40", "cachedInput": "0.14", "sourceUrl": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07", "notes": ["New entry: absent from this page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than this file's conservative 2026-01-01.", `Listed under "Third-Party Models" on Mistral's own pricing page - a Z.ai GLM model resold through La Plateforme, so this is Mistral's resale rate, not Z.ai's first-party rate.`, "All three rates are printed per-model on the page. No batch discount is stated for the third-party section, so batchMultiplier is omitted rather than assumed from the first-party models' 50%."] }
+    ],
+    source: { "url": "https://mistral.ai/pricing/api", "observedAt": "2026-09-07" }
+  },
   {
     canonicalId: "gpt-3.5-turbo",
     provider: "openai",
     aliases: [],
     family: "gpt-3.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Found on https://developers.openai.com/api/docs/pricing during a second confirmation pass.", "No cached-input rate is published for gpt-3.5-turbo on the current pricing page; cachedInput is intentionally omitted rather than guessed.", "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "1.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Found on https://developers.openai.com/api/docs/pricing during a second confirmation pass.", "No cached-input rate is published for gpt-3.5-turbo on the current pricing page; cachedInput is intentionally omitted rather than guessed.", "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-4.1",
@@ -1087,9 +1276,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-4.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "8.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "8.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-4.1-mini",
@@ -1097,9 +1286,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-4.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.40", "output": "1.60", "cachedInput": "0.10", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.40", "output": "1.60", "cachedInput": "0.10", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-4.1-nano",
@@ -1107,9 +1296,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-4.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.40", "cachedInput": "0.025", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.40", "cachedInput": "0.025", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-4o",
@@ -1117,9 +1306,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-4o",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "10.00", "cachedInput": "1.25", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "10.00", "cachedInput": "1.25", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-4o-mini",
@@ -1127,9 +1316,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-4o",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5",
@@ -1137,9 +1326,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model.", "Lead-supplied verified table (observed 2026-08-05) matches this rate exactly; independently re-confirmed against https://developers.openai.com/api/docs/pricing on the same date."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model.", "Lead-supplied verified table (observed 2026-08-05) matches this rate exactly; independently re-confirmed against https://developers.openai.com/api/docs/pricing on the same date."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5-mini",
@@ -1147,9 +1336,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.25", "output": "2.00", "cachedInput": "0.025", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.25", "output": "2.00", "cachedInput": "0.025", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5-nano",
@@ -1157,9 +1346,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.40", "cachedInput": "0.005", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.40", "cachedInput": "0.005", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5-pro",
@@ -1167,9 +1356,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "120.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ['No cached-input rate is published for gpt-5-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier is intentionally omitted for this pro-tier model; not independently confirmed to apply."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "120.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ['No cached-input rate is published for gpt-5-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.1",
@@ -1177,9 +1366,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.1",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.2",
@@ -1187,9 +1376,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.2",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.75", "output": "14.00", "cachedInput": "0.175", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.75", "output": "14.00", "cachedInput": "0.175", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.2-pro",
@@ -1197,9 +1386,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.2",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "21.00", "output": "168.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ['No cached-input rate is published for gpt-5.2-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier is intentionally omitted for this pro-tier model; not independently confirmed to apply."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "21.00", "output": "168.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ['No cached-input rate is published for gpt-5.2-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.4",
@@ -1207,9 +1396,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.4",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "15.00", "cachedInput": "0.25", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.50", "output": "15.00", "cachedInput": "0.25", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model.", 'The pricing page qualifies this rate as the "<272K" context tier. A higher tier above 272K tokens is implied by that label but no rate for it was published in the fetched table, and this schema has no context-length dimension, so only the <272K rate is recorded.'] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.4-mini",
@@ -1217,9 +1406,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.4",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.75", "output": "4.50", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.75", "output": "4.50", "cachedInput": "0.075", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.4-nano",
@@ -1227,9 +1416,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.4",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "1.25", "cachedInput": "0.02", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "1.25", "cachedInput": "0.02", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.4-pro",
@@ -1237,9 +1426,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.4",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "30.00", "output": "180.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ['No cached-input rate is published for gpt-5.4-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier is intentionally omitted for this pro-tier model; not independently confirmed to apply."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "30.00", "output": "180.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ['No cached-input rate is published for gpt-5.4-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`, 'The pricing page qualifies this rate as the "<272K" context tier. A higher tier above 272K tokens is implied by that label but no rate for it was published in the fetched table, and this schema has no context-length dimension, so only the <272K rate is recorded.'] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.5",
@@ -1247,9 +1436,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model.", 'The pricing page qualifies this rate as the "<272K" context tier. A higher tier above 272K tokens is implied by that label but no rate for it was published in the fetched table, and this schema has no context-length dimension, so only the <272K rate is recorded.'] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.5-pro",
@@ -1257,9 +1446,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.5",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "30.00", "output": "180.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ['No cached-input rate is published for gpt-5.5-pro (shown as "\u2014" on the pricing page); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is intentionally omitted for this pro-tier model: the pricing page's blanket "50% off Batch" statement was not independently confirmed to apply to the -pro tier, unlike the base tiers.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "30.00", "output": "180.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ['No cached-input rate is published for gpt-5.5-pro (shown as "\u2014" on the pricing page); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`, 'The pricing page qualifies this rate as the "<272K" context tier. A higher tier above 272K tokens is implied by that label but no rate for it was published in the fetched table, and this schema has no context-length dimension, so only the <272K rate is recorded.'] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.6-luna",
@@ -1267,9 +1456,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.6",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "1.20", "cachedInput": "0.02", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.20", "output": "1.20", "cachedInput": "0.02", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.6-sol",
@@ -1277,9 +1466,10 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.6",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": [`OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date (this model's naming implies a later release, but a conservative too-early effectiveFrom only ever makes a historical lookup succeed when it should return "no period found", never the reverse).`, `batchMultiplier reflects OpenAI's general Batch API policy ("a 50% discount to Standard pricing rates across all models") as stated on the pricing page; not independently confirmed per-model.`] }
+      { "effectiveFrom": "2026-01-01", "effectiveTo": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "5.00", "output": "30.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": [`OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date (this model's naming implies a later release, but a conservative too-early effectiveFrom only ever makes a historical lookup succeed when it should return "no period found", never the reverse).`, `batchMultiplier reflects OpenAI's general Batch API policy ("a 50% discount to Standard pricing rates across all models") as stated on the pricing page; not independently confirmed per-model.`, "Closed on 2026-09-07: the pricing page published a lower rate (4.00 input / 20.00 output) on that date. The old rate was last confirmed 2026-08-05, so the true change date lies in (2026-08-05, 2026-09-07]; effectiveTo is the observation date, which keeps every confirmed observation correct and approximates only the unobserved gap, toward the last confirmed value."] },
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "4.00", "output": "20.00", "cachedInput": "0.40", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Price cut observed on 2026-09-07: input 5.00 -> 4.00, output 30.00 -> 20.00, cached input 0.50 -> 0.40. OpenAI publishes no effective date, so effectiveFrom is the observation date rather than a guess at when the cut actually landed.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-5.6-terra",
@@ -1287,9 +1477,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-5.6",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cachedInput": "0.20", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cachedInput": "0.20", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+    ],
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "gpt-6-astra",
+    provider: "openai",
+    aliases: [],
+    family: "gpt-6",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "10.00", "output": "50.00", "cachedInput": "1.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["New model: absent from the pricing page on 2026-08-05, present on 2026-09-07. effectiveFrom is the observation date rather than this file's usual conservative 2026-01-01 - the model demonstrably did not exist at that rate a month earlier, so backdating it would invent a period.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o1",
@@ -1297,9 +1497,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "60.00", "cachedInput": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "15.00", "output": "60.00", "cachedInput": "7.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o1-pro",
@@ -1307,9 +1507,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "150.00", "output": "600.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["Found on https://developers.openai.com/api/docs/pricing during a second confirmation pass.", 'No cached-input rate is published for o1-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier is intentionally omitted for this pro-tier model; not independently confirmed to apply."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "150.00", "output": "600.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["Found on https://developers.openai.com/api/docs/pricing during a second confirmation pass.", 'No cached-input rate is published for o1-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o3",
@@ -1317,9 +1517,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "8.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "8.00", "cachedInput": "0.50", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o3-mini",
@@ -1327,9 +1527,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.10", "output": "4.40", "cachedInput": "0.55", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.10", "output": "4.40", "cachedInput": "0.55", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o3-pro",
@@ -1337,9 +1537,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "20.00", "output": "80.00", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ['No cached-input rate is published for o3-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier is intentionally omitted for this pro-tier model; not independently confirmed to apply."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "20.00", "output": "80.00", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ['No cached-input rate is published for o3-pro (shown as "\u2014"); cachedInput is intentionally omitted rather than guessed.', "OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", `batchMultiplier is "0.5" per the pricing page's Batch API section, which states the 50% saving "applies to all text models listed above in the Standard table" - an explicit all-model statement, so the pro tiers are covered too.`] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "o4-mini",
@@ -1347,9 +1547,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "o-series",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.10", "output": "4.40", "cachedInput": "0.275", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.10", "output": "4.40", "cachedInput": "0.275", "batchMultiplier": "0.5", "sourceUrl": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07", "notes": ["OpenAI's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01 pending confirmation of the true rollout date.", "batchMultiplier reflects OpenAI's general Batch API policy as stated on the pricing page; not independently confirmed per-model."] }
     ],
-    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://developers.openai.com/api/docs/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "anthropic/claude-sonnet-5",
@@ -1357,9 +1557,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "anthropic-proxy",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "10.00", "sourceUrl": "https://openrouter.ai/anthropic/claude-sonnet-5", "observedAt": "2026-08-05", "notes": ["UNCERTAIN \u2014 flagged explicitly: this rate ($2.00/$10.00) matches Anthropic's own INTRODUCTORY rate for claude-sonnet-5, which anthropic.json records as expiring 2026-08-31 and being replaced by a $3.00/$15.00 standard rate from 2026-09-01 (see anthropic.json). It is not clear from the OpenRouter page alone whether OpenRouter (a) has simply not yet updated its listing to the post-introductory rate, (b) is genuinely offering a different long-term rate than Anthropic's own API, or (c) this reflects a caching/rounding artifact in the page. Recorded as fetched and observed on 2026-08-05, but a follow-up reviewer should re-check this specific model close to and after 2026-09-01.", "OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", `canonicalId uses OpenRouter's own slug format, deliberately distinct from Anthropic's first-party canonicalId "claude-sonnet-5" (anthropic.json) to avoid a canonicalId collision; this is a legitimate cross-provider situation, not an error.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "10.00", "cachedInput": "0.20", "cacheWrite": "2.50", "sourceUrl": "https://openrouter.ai/anthropic/claude-sonnet-5", "observedAt": "2026-09-07", "notes": ["Recorded 2026-08-05 flagged UNCERTAIN: $2.00/$10.00 matched what anthropic.json then held as an introductory rate expiring 2026-08-31, so it was unclear whether OpenRouter had simply not updated its listing. Resolved on 2026-09-07 - Anthropic's pricing page states the scheduled $3.00/$15.00 increase will not occur and $2.00/$10.00 is the standard rate, so this listing was correct all along and the flag is withdrawn.", "OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", `canonicalId uses OpenRouter's own slug format, deliberately distinct from Anthropic's first-party canonicalId "claude-sonnet-5" (anthropic.json) to avoid a canonicalId collision; this is a legitimate cross-provider situation, not an error.`, "Re-confirmed unchanged on 2026-09-07 against the same model page.", "cachedInput added on 2026-09-07: the model page now prints a Cache Read rate of 0.20 per million tokens for this model.", "cacheWrite added on 2026-09-07: the model page prints a 5-minute Cache Write rate of 2.50 per million tokens (and $4.00 for the 1-hour TTL, which this single-field schema does not model).", "Resolves the 2026-08-05 caveat on this entry: $2.00/$10.00 was flagged as possibly Anthropic's introductory rate, due to be superseded by $3.00/$15.00 on 2026-09-01. Anthropic's own pricing page now states that increase will not occur and $2.00/$10.00 is the standard rate, so OpenRouter's rate matches the first-party standard rate, not a stale introductory one.", "The page notes Google Vertex (US/Europe) and Amazon Bedrock (US) upstreams charge $2.20/$11.00 through OpenRouter; the default cross-provider rate is recorded, since this schema has no upstream dimension."] }
     ],
-    source: { "url": "https://openrouter.ai/anthropic/claude-sonnet-5", "observedAt": "2026-08-05" }
+    source: { "url": "https://openrouter.ai/anthropic/claude-sonnet-5", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "google/gemini-3.1-pro-preview",
@@ -1367,9 +1567,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "google-proxy",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cheapestTier": true, "sourceUrl": "https://openrouter.ai/google/gemini-3.1-pro-preview", "observedAt": "2026-08-05", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", "Matches Google's own first-party <=200k-token-tier rate for gemini-3.1-pro-preview ($2.00/$12.00, see google.json) exactly as observed. Google's >200k-token tier ($4.00/$18.00) is not represented here (or, evidently, distinguished by OpenRouter's listing either) \u2014 same context-length-tiering limitation as google.json.", `canonicalId uses OpenRouter's own slug format, deliberately distinct from Google's first-party canonicalId "gemini-3.1-pro-preview" (google.json).`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "12.00", "cachedInput": "0.20", "cheapestTier": true, "sourceUrl": "https://openrouter.ai/google/gemini-3.1-pro-preview", "observedAt": "2026-09-07", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", "Matches Google's own first-party <=200k-token-tier rate for gemini-3.1-pro-preview ($2.00/$12.00, see google.json) exactly as observed. Google's >200k-token tier ($4.00/$18.00) is not represented here (or, evidently, distinguished by OpenRouter's listing either) \u2014 same context-length-tiering limitation as google.json.", `canonicalId uses OpenRouter's own slug format, deliberately distinct from Google's first-party canonicalId "gemini-3.1-pro-preview" (google.json).`, "Re-confirmed unchanged on 2026-09-07 against the same model page.", "cachedInput added on 2026-09-07: the model page now prints a Cache Read rate of 0.20 per million tokens for this model."] }
     ],
-    source: { "url": "https://openrouter.ai/google/gemini-3.1-pro-preview", "observedAt": "2026-08-05" }
+    source: { "url": "https://openrouter.ai/google/gemini-3.1-pro-preview", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "meta-llama/llama-3.3-70b-instruct",
@@ -1377,9 +1577,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "meta-proxy",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.32", "sourceUrl": "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct", "observedAt": "2026-08-05", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", `Meta does not sell first-party API access to Llama, so there is no first-party "llama-3.3-70b" entry in this registry to compare against; Groq (groq.json: llama-3.3-70b-versatile, $0.59/$0.79) and Together AI (together.json: llama-3.3-70b, $1.04/$1.04) each host the same open-weight model at their own, different rates. OpenRouter's rate here is the lowest of the three observed, plausibly because OpenRouter itself proxies to one of several underlying hosts and shows a blended/lowest-cost route; not independently confirmed which underlying host this routes to.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.10", "output": "0.32", "sourceUrl": "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct", "observedAt": "2026-09-07", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", `Meta does not sell first-party API access to Llama, so there is no first-party "llama-3.3-70b" entry in this registry to compare against; Groq (groq.json: llama-3.3-70b-versatile, $0.59/$0.79) and Together AI (together.json: llama-3.3-70b, $1.04/$1.04) each host the same open-weight model at their own, different rates. OpenRouter's rate here is the lowest of the three observed, plausibly because OpenRouter itself proxies to one of several underlying hosts and shows a blended/lowest-cost route; not independently confirmed which underlying host this routes to.`, "Re-confirmed unchanged on 2026-09-07 against the same model page.", 'The page labels this "the average price customers actually pay" and warns caching and discounts often put the effective price below it; recorded as printed.'] }
     ],
-    source: { "url": "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct", "observedAt": "2026-08-05" }
+    source: { "url": "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "openai/gpt-5",
@@ -1387,9 +1587,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "openai-proxy",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "sourceUrl": "https://openrouter.ai/openai/gpt-5", "observedAt": "2026-08-05", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", "Matches OpenAI's own first-party rate for gpt-5 ($1.25/$10.00, see openai.json) exactly as observed \u2014 no markup detected for this model.", `canonicalId uses OpenRouter's own slug format ("openai/gpt-5"), deliberately distinct from OpenAI's first-party canonicalId "gpt-5" (openai.json) \u2014 this avoids a canonicalId collision while still allowing a cross-provider alias collision if a caller looks up the bare id "gpt-5" without a provider qualifier; no bare "gpt-5" alias was added to this entry to keep that surface area minimal.`] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "10.00", "cachedInput": "0.125", "sourceUrl": "https://openrouter.ai/openai/gpt-5", "observedAt": "2026-09-07", "notes": ["OpenRouter's per-model page does not publish an effective date; effectiveFrom is set conservatively to 2026-01-01.", "Matches OpenAI's own first-party rate for gpt-5 ($1.25/$10.00, see openai.json) exactly as observed \u2014 no markup detected for this model.", `canonicalId uses OpenRouter's own slug format ("openai/gpt-5"), deliberately distinct from OpenAI's first-party canonicalId "gpt-5" (openai.json) \u2014 this avoids a canonicalId collision while still allowing a cross-provider alias collision if a caller looks up the bare id "gpt-5" without a provider qualifier; no bare "gpt-5" alias was added to this entry to keep that surface area minimal.`, "Re-confirmed unchanged on 2026-09-07 against the same model page.", "cachedInput added on 2026-09-07: the model page now prints a Cache Read rate of 0.125 per million tokens for this model."] }
+    ],
+    source: { "url": "https://openrouter.ai/openai/gpt-5", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "deepseek-v4-flash-0731",
+    provider: "together",
+    aliases: [],
+    family: "deepseek",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.14", "output": "0.28", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
     ],
-    source: { "url": "https://openrouter.ai/openai/gpt-5", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "deepseek-v4-pro",
@@ -1397,19 +1607,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "deepseek",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.74", "output": "3.48", "cachedInput": "0.20", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.74", "output": "3.48", "cachedInput": "0.20", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Not re-confirmed on 2026-09-07: that fetch listed "DeepSeek V4 Pro 0813" at $1.32 / $3.96 and no undated "DeepSeek V4 Pro" row. Whether the dated build is this same model repriced or a separate snapshot is not stated on the page, so this entry keeps its 2026-08-05 rate and the dated build is recorded separately as deepseek-v4-pro-0813 rather than silently overwriting this one.'] }
     ],
     source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
   },
+  {
+    canonicalId: "deepseek-v4-pro-0813",
+    provider: "together",
+    aliases: [],
+    family: "deepseek",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "1.32", "output": "3.96", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Dated build listed on the page on 2026-09-07; see deepseek-v4-pro's notes for why it is a separate entry rather than a reprice of that one.", "Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
   {
     canonicalId: "gemma-4-31b",
     provider: "together",
     aliases: [],
     family: "gemma",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.39", "output": "0.97", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): AWS Bedrock also lists "Gemma 4 31B" (aws-bedrock.json: gemma-4-31b) at a different, lower rate ($0.14/$0.40 as observed on Bedrock) \u2014 same underlying Google open-weight model, independently priced by each reseller; do not assume parity.'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.39", "output": "0.97", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): AWS Bedrock also lists "Gemma 4 31B" (aws-bedrock.json: gemma-4-31b) at a different, lower rate ($0.14/$0.40 as observed on Bedrock) \u2014 same underlying Google open-weight model, independently priced by each reseller; do not assume parity.', "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "glm-5.2",
@@ -1417,9 +1637,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "glm",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.40", "output": "4.40", "cachedInput": "0.26", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.40", "output": "4.40", "cachedInput": "0.26", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "glm-5.3",
+    provider: "together",
+    aliases: [],
+    family: "glm",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "1.40", "output": "4.40", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "glm-5.3-flash",
+    provider: "together",
+    aliases: [],
+    family: "glm",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.50", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-oss-120b",
@@ -1427,9 +1667,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-oss",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Groq also publishes a model with canonicalId "gpt-oss-120b" (groq.json), independently priced at the same $0.15/$0.60 figure as observed. The identical string "gpt-oss-120b" is used as the canonicalId on both providers because that is the actual model name each provider publishes; resolving "gpt-oss-120b" without a provider qualifier is ambiguous across providers by design and the resolver requires a provider qualifier to disambiguate it.'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.60", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Groq also publishes a model with canonicalId "gpt-oss-120b" (groq.json), independently priced at the same $0.15/$0.60 figure as observed. The identical string "gpt-oss-120b" is used as the canonicalId on both providers because that is the actual model name each provider publishes; resolving "gpt-oss-120b" without a provider qualifier is ambiguous across providers by design and the resolver requires a provider qualifier to disambiguate it.', "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "gpt-oss-20b",
@@ -1437,7 +1677,7 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "gpt-oss",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.20", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Groq also publishes "gpt-oss-20b" (groq.json) at a different rate ($0.075/$0.30 as observed) \u2014 same model name, independently priced by each host; do not assume parity.'] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.05", "output": "0.20", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", 'Cross-provider alias/canonicalId collision (allowed, not an error): Groq also publishes "gpt-oss-20b" (groq.json) at a different rate ($0.075/$0.30 as observed) \u2014 same model name, independently priced by each host; do not assume parity.', "Not surfaced by the 2026-09-07 fetch of the same page. observedAt is deliberately left at 2026-08-05: the rate was not re-observed, and an absence in one page fetch is not evidence of a withdrawal."] }
     ],
     source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
   },
@@ -1447,9 +1687,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "kimi",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "3.00", "output": "15.00", "cachedInput": "0.30", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "llama-3-8b-instruct-lite",
+    provider: "together",
+    aliases: [],
+    family: "llama",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.14", "output": "0.14", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "llama-3.3-70b",
@@ -1457,9 +1707,9 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "llama-3.3",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.04", "output": "1.04", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Also hosted by Groq (groq.json: llama-3.3-70b-versatile, $0.59/$0.79) and resold by AWS Bedrock \u2014 each host prices this same open-weight model independently; do not assume parity across providers."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.04", "output": "1.04", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Also hosted by Groq (groq.json: llama-3.3-70b-versatile, $0.59/$0.79) and resold by AWS Bedrock \u2014 each host prices this same open-weight model independently; do not assume parity across providers.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "minimax-m3",
@@ -1467,9 +1717,19 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "minimax",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "1.20", "cachedInput": "0.06", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "1.20", "cachedInput": "0.06", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen2.5-7b-instruct-turbo",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.30", "output": "0.30", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "qwen3.5-397b-a17b",
@@ -1477,9 +1737,29 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "qwen",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "3.60", "cachedInput": "0.35", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "0.60", "output": "3.60", "cachedInput": "0.35", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.5-9b",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.17", "output": "0.25", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.6-plus",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.50", "output": "3.00", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   },
   {
     canonicalId: "qwen3.7-max",
@@ -1487,9 +1767,39 @@ var MODEL_REGISTRY = [
     aliases: [],
     family: "qwen",
     pricing: [
-      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "3.75", "cachedInput": "0.13", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-08-05", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01."] }
+      { "effectiveFrom": "2026-01-01", "currency": "USD", "unit": "per-million-tokens", "input": "1.25", "output": "3.75", "cachedInput": "0.13", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Together's pricing page does not publish an effective date for this rate; effectiveFrom is set conservatively to 2026-01-01.", "Re-confirmed unchanged on 2026-09-07 against the same page."] }
     ],
-    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-08-05" }
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.7-plus",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.32", "output": "1.28", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.8-2.4t-a95b",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "2.00", "output": "6.00", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
+  },
+  {
+    canonicalId: "qwen3.8-flash",
+    provider: "together",
+    aliases: [],
+    family: "qwen",
+    pricing: [
+      { "effectiveFrom": "2026-09-07", "currency": "USD", "unit": "per-million-tokens", "input": "0.15", "output": "0.47", "sourceUrl": "https://www.together.ai/pricing", "observedAt": "2026-09-07", "notes": ["Added from the 2026-09-07 fetch. effectiveFrom is the observation date rather than this file's conservative 2026-01-01, since the 2026-08-05 fetch of the same page did not list it.", "Together publishes no cached-input rate or batch discount for serverless models; both fields are omitted rather than guessed."] }
+    ],
+    source: { "url": "https://www.together.ai/pricing", "observedAt": "2026-09-07" }
   }
 ];
 
@@ -1549,7 +1859,7 @@ var InvalidRateError = class extends Error {
 var ZERO = { numerator: 0n, scale: 0 };
 var DECIMAL_PATTERN = /^\d+(\.\d+)?$/;
 function parseDecimalRate(value, field) {
-  if (!DECIMAL_PATTERN.test(value)) {
+  if (typeof value !== "string" || !DECIMAL_PATTERN.test(value)) {
     throw new InvalidRateError(field, value);
   }
   const dot = value.indexOf(".");
@@ -1983,6 +2293,59 @@ function createPriceOverride(input) {
   };
 }
 
+// src/aggregate.ts
+function sumExactUsd(values) {
+  const amounts = [];
+  for (const [index, value] of values.entries()) {
+    amounts.push(parseDecimalRate(value, `values[${String(index)}]`));
+  }
+  return formatExact(addExact(amounts));
+}
+function modelKey(breakdown) {
+  return `${breakdown.provider}:${breakdown.canonicalModel}`;
+}
+function newBucket() {
+  return { count: 0, sum: ZERO, registryVersions: /* @__PURE__ */ new Set() };
+}
+function accumulate(bucket, amount, registryVersion) {
+  bucket.count += 1;
+  bucket.sum = addExact([bucket.sum, amount]);
+  bucket.registryVersions.add(registryVersion);
+}
+function toTotal(bucket) {
+  return {
+    count: bucket.count,
+    totalUsd: toDisplayNumber(bucket.sum),
+    totalUsdExact: formatExact(bucket.sum),
+    registryVersions: [...bucket.registryVersions].sort()
+  };
+}
+function createCostAggregator() {
+  const overall = newBucket();
+  const byModelBuckets = /* @__PURE__ */ new Map();
+  return {
+    add(breakdown) {
+      const amount = parseDecimalRate(breakdown.totalUsdExact, "totalUsdExact");
+      const key = modelKey(breakdown);
+      let bucket = byModelBuckets.get(key);
+      if (bucket === void 0) {
+        bucket = newBucket();
+        byModelBuckets.set(key, bucket);
+      }
+      accumulate(bucket, amount, breakdown.registryVersion);
+      accumulate(overall, amount, breakdown.registryVersion);
+    },
+    total() {
+      return toTotal(overall);
+    },
+    byModel() {
+      const result = /* @__PURE__ */ new Map();
+      for (const [key, bucket] of byModelBuckets) result.set(key, toTotal(bucket));
+      return result;
+    }
+  };
+}
+
 // src/normalize/openai.ts
 var KNOWN_TOP_LEVEL = /* @__PURE__ */ new Set([
   "prompt_tokens",