language-models 2.1.3 → 2.3.0

package/dist/pricing/lookup.js

@@ -0,0 +1,94 @@
+/**
+ * language-models / pricing — lookup helpers.
+ *
+ * The single source of truth for cost computation across the Phase-2
+ * three-repo cascade. Throws on unknown slug/tier (per BYOK_GATEWAY_LIES
+ * memory: silent zero is the lying-gateway pattern; loud failure beats
+ * silent downgrade).
+ */
+import { PRICING_TABLE } from './table.js';
+/**
+ * Compute USD cost for a known generation. Throws on:
+ * - unknown slug (no rows for the slug at all)
+ * - unknown/unmodeled tier for the slug (no row for the (slug, tier) pair)
+ * - negative token counts (programming error — fail loud)
+ *
+ * Honors `contextTierBreakpoint` when present: if `inputTokens >=
+ * breakpoint`, the entire request is billed at `contextTierAbove` rates
+ * (matches Google's published billing model — the rate switches once the
+ * input crosses 200K, applied to the full request).
+ *
+ * `cachedInputTokens` is billed at `cachedInputPer1M` when defined,
+ * falling back to the regular input rate otherwise. Cached tokens are
+ * SUBTRACTED from `inputTokens` to compute the non-cached portion — i.e.
+ * `inputTokens` is the TOTAL input including any cached tokens.
+ */
+export function priceFor(args) {
+    const { slug, tier, inputTokens, outputTokens, cachedInputTokens } = args;
+    if (inputTokens < 0 || outputTokens < 0 || (cachedInputTokens ?? 0) < 0) {
+        throw new RangeError(`priceFor() requires non-negative token counts; got input=${inputTokens}, output=${outputTokens}, cachedInput=${cachedInputTokens ?? 0}`);
+    }
+    const row = findRow(slug, tier);
+    // Pick the right rate block based on context-tier breakpoint.
+    const useAbove = typeof row.contextTierBreakpoint === 'number' &&
+        row.contextTierAbove !== undefined &&
+        inputTokens >= row.contextTierBreakpoint;
+    const block = useAbove
+        ? row.contextTierAbove
+        : {
+            inputPer1M: row.inputPer1M,
+            outputPer1M: row.outputPer1M,
+            cachedInputPer1M: row.cachedInputPer1M,
+        };
+    const cachedTok = Math.min(cachedInputTokens ?? 0, inputTokens);
+    const nonCachedInputTok = inputTokens - cachedTok;
+    const cachedRate = block.cachedInputPer1M ?? block.inputPer1M;
+    const inputUsd = (nonCachedInputTok / 1_000_000) * block.inputPer1M + (cachedTok / 1_000_000) * cachedRate;
+    const outputUsd = (outputTokens / 1_000_000) * block.outputPer1M;
+    const totalUsd = inputUsd + outputUsd;
+    return { inputUsd, outputUsd, totalUsd };
+}
+/**
+ * Returns the unique set of slugs in the table. Useful for adapters that
+ * want to validate caller-supplied model ids before dispatching.
+ */
+export function listSlugs() {
+    const set = new Set();
+    for (const row of PRICING_TABLE)
+        set.add(row.slug);
+    return Array.from(set);
+}
+/**
+ * Returns true if pricing exists for the given (slug, tier). Use this
+ * when you need a non-throwing existence check before calling
+ * `priceFor()` (e.g. a metering middleware that wants to fall back to
+ * "unknown cost" telemetry rather than throwing on a not-yet-registered
+ * model).
+ */
+export function hasPricing(args) {
+    return PRICING_TABLE.some((row) => row.slug === args.slug && row.tier === args.tier);
+}
+/**
+ * Returns all pricing rows for a slug, across all tiers. Useful for
+ * tooling that wants to display "this model has standard + batch tiers"
+ * in a UI.
+ */
+export function rowsForSlug(slug) {
+    return PRICING_TABLE.filter((row) => row.slug === slug);
+}
+// ---------------------------------------------------------------------------
+// Internal
+// ---------------------------------------------------------------------------
+function findRow(slug, tier) {
+    const slugRows = PRICING_TABLE.filter((row) => row.slug === slug);
+    if (slugRows.length === 0) {
+        throw new Error(`Unknown model slug: '${slug}'. Known slugs: ${listSlugs().slice(0, 8).join(', ')}${listSlugs().length > 8 ? `, ...(${listSlugs().length - 8} more)` : ''}`);
+    }
+    const tierRow = slugRows.find((row) => row.tier === tier);
+    if (tierRow === undefined) {
+        const availableTiers = slugRows.map((row) => row.tier);
+        throw new Error(`No '${tier}' pricing for slug '${slug}'. Available tiers: ${availableTiers.join(', ')}`);
+    }
+    return tierRow;
+}
+//# sourceMappingURL=lookup.js.map

package/dist/pricing/lookup.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"lookup.js","sourceRoot":"","sources":["../../src/pricing/lookup.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAA;AAS1C;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,QAAQ,CAAC,IAAkB;IACzC,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,YAAY,EAAE,iBAAiB,EAAE,GAAG,IAAI,CAAA;IAEzE,IAAI,WAAW,GAAG,CAAC,IAAI,YAAY,GAAG,CAAC,IAAI,CAAC,iBAAiB,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QACxE,MAAM,IAAI,UAAU,CAClB,4DAA4D,WAAW,YAAY,YAAY,iBAC7F,iBAAiB,IAAI,CACvB,EAAE,CACH,CAAA;IACH,CAAC;IAED,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAA;IAE/B,8DAA8D;IAC9D,MAAM,QAAQ,GACZ,OAAO,GAAG,CAAC,qBAAqB,KAAK,QAAQ;QAC7C,GAAG,CAAC,gBAAgB,KAAK,SAAS;QAClC,WAAW,IAAI,GAAG,CAAC,qBAAqB,CAAA;IAC1C,MAAM,KAAK,GAAG,QAAQ;QACpB,CAAC,CAAE,GAAG,CAAC,gBAA6D;QACpE,CAAC,CAAC;YACE,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,gBAAgB,EAAE,GAAG,CAAC,gBAAgB;SACvC,CAAA;IAEL,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,iBAAiB,IAAI,CAAC,EAAE,WAAW,CAAC,CAAA;IAC/D,MAAM,iBAAiB,GAAG,WAAW,GAAG,SAAS,CAAA;IACjD,MAAM,UAAU,GAAG,KAAK,CAAC,gBAAgB,IAAI,KAAK,CAAC,UAAU,CAAA;IAE7D,MAAM,QAAQ,GACZ,CAAC,iBAAiB,GAAG,SAAS,CAAC,GAAG,KAAK,CAAC,UAAU,GAAG,CAAC,SAAS,GAAG,SAAS,CAAC,GAAG,UAAU,CAAA;IAC3F,MAAM,SAAS,GAAG,CAAC,YAAY,GAAG,SAAS,CAAC,GAAG,KAAK,CAAC,WAAW,CAAA;IAChE,MAAM,QAAQ,GAAG,QAAQ,GAAG,SAAS,CAAA;IAErC,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAA;AAC1C,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,SAAS;IACvB,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAA;IAC7B,KAAK,MAAM,GAAG,IAAI,aAAa;QAAE,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;IAClD,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;AACxB,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,UAAU,CAAC,IAAoB;IAC7C,OAAO,aAAa,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI,CAAC,CAAA;AACtF,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,WAAW,CAAC,IAAY;IACtC,OAAO,aAAa,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI,CAAC,CAAA;AACzD,CAAC;AAED,8EAA8E;AAC9E,WAAW;AACX,8EAA8E;AAE9E,SAAS,OAAO,CAAC,IAAY,EAAE,IAAiB;IAC9C,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI,CAAC,CAAA;IACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,mBAAmB,SAAS,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAC/E,SAAS,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,SAAS,EAAE,CAAC,MAAM,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EACrE,EAAE,CACH,CAAA;IACH,CAAC;IACD,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI,CAAC,CAAA;IACzD,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,cAAc,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;QACtD,MAAM,IAAI,KAAK,CACb,OAAO,IAAI,uBAAuB,IAAI,uBAAuB,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACzF,CAAA;IACH,CAAC;IACD,OAAO,OAAO,CAAA;AAChB,CAAC"}

package/dist/pricing/table.d.ts

@@ -0,0 +1,46 @@
+/**
+ * language-models / pricing — canonical model pricing table.
+ *
+ * All rates are USD per 1,000,000 tokens (per1M form, NOT per1k). Sourced
+ * from public list prices on 2026-05-07. Where prior repo tables disagree,
+ * the most-recently-updated source wins:
+ *
+ * - **Vertex Gemini 3.x**: startup-builder/packages/llm-vertex-batch
+ *   (sb-srnl 2026-05-07 verified) is canonical for the flex/batch tier.
+ *   startup-builder/packages/llm-vertex (same date) is canonical for
+ *   standard interactive pricing.
+ * - **Vertex Gemini 2.5**: startup-builder/llm-vertex-batch is canonical
+ *   (icps's `gemini-2.5-flash` rates were ~4× off — likely transposed
+ *   from a different SKU).
+ * - **Bedrock Anthropic**: startup-builder/packages/llm-bedrock is the
+ *   most-recently-curated table (matches AWS Bedrock public pricing on
+ *   2026-05-07).
+ * - **Embeddings**: icps/packages/llm/pricing.ts is the source for
+ *   `gemini-embedding-2`. Bedrock has no first-party Gemini embedding
+ *   SKU; Anthropic has no embedding SKU. Embedding lives under
+ *   `aistudio/` since the cheapest path is the AI Studio API key (icps's
+ *   2026-05-07 fallback chain — the path startup-builder also uses).
+ *
+ * 200K-context tier breakpoint:
+ *
+ * Gemini 3.1 Pro Preview list pricing has a 2× rate above 200K input
+ * tokens. We model this with `contextTierBreakpoint: 200_000` +
+ * `contextTierAbove`. Anthropic Claude has flat pricing (no breakpoint).
+ * Gemini 2.5 + 3.x flash-lite + embedding SKUs are flat too.
+ *
+ * Adding a new SKU:
+ *
+ * 1. Append a new row keyed on `<provider>/<short-slug>` + `tier`. Both
+ *    fields together form the lookup key; duplicates are caught by the
+ *    table-integrity test.
+ * 2. If the SKU has a context-tier breakpoint, add
+ *    `contextTierBreakpoint` + `contextTierAbove`. Otherwise omit them.
+ * 3. If both standard + batch tiers exist, add BOTH rows; downstream
+ *    consumers select via `priceFor({ tier: ... })`. Vertex's batch tier
+ *    is ~50% of standard; that's not enforced — declare the actual
+ *    published rates.
+ * 4. Bump the package version (semver minor for additions).
+ */
+import type { ModelPricing } from './types.js';
+export declare const PRICING_TABLE: readonly ModelPricing[];
+//# sourceMappingURL=table.d.ts.map

package/dist/pricing/table.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"table.d.ts","sourceRoot":"","sources":["../../src/pricing/table.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA0CG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAE9C,eAAO,MAAM,aAAa,EAAE,SAAS,YAAY,EA4LhD,CAAA"}

package/dist/pricing/table.js

@@ -0,0 +1,214 @@
+/**
+ * language-models / pricing — canonical model pricing table.
+ *
+ * All rates are USD per 1,000,000 tokens (per1M form, NOT per1k). Sourced
+ * from public list prices on 2026-05-07. Where prior repo tables disagree,
+ * the most-recently-updated source wins:
+ *
+ * - **Vertex Gemini 3.x**: startup-builder/packages/llm-vertex-batch
+ *   (sb-srnl 2026-05-07 verified) is canonical for the flex/batch tier.
+ *   startup-builder/packages/llm-vertex (same date) is canonical for
+ *   standard interactive pricing.
+ * - **Vertex Gemini 2.5**: startup-builder/llm-vertex-batch is canonical
+ *   (icps's `gemini-2.5-flash` rates were ~4× off — likely transposed
+ *   from a different SKU).
+ * - **Bedrock Anthropic**: startup-builder/packages/llm-bedrock is the
+ *   most-recently-curated table (matches AWS Bedrock public pricing on
+ *   2026-05-07).
+ * - **Embeddings**: icps/packages/llm/pricing.ts is the source for
+ *   `gemini-embedding-2`. Bedrock has no first-party Gemini embedding
+ *   SKU; Anthropic has no embedding SKU. Embedding lives under
+ *   `aistudio/` since the cheapest path is the AI Studio API key (icps's
+ *   2026-05-07 fallback chain — the path startup-builder also uses).
+ *
+ * 200K-context tier breakpoint:
+ *
+ * Gemini 3.1 Pro Preview list pricing has a 2× rate above 200K input
+ * tokens. We model this with `contextTierBreakpoint: 200_000` +
+ * `contextTierAbove`. Anthropic Claude has flat pricing (no breakpoint).
+ * Gemini 2.5 + 3.x flash-lite + embedding SKUs are flat too.
+ *
+ * Adding a new SKU:
+ *
+ * 1. Append a new row keyed on `<provider>/<short-slug>` + `tier`. Both
+ *    fields together form the lookup key; duplicates are caught by the
+ *    table-integrity test.
+ * 2. If the SKU has a context-tier breakpoint, add
+ *    `contextTierBreakpoint` + `contextTierAbove`. Otherwise omit them.
+ * 3. If both standard + batch tiers exist, add BOTH rows; downstream
+ *    consumers select via `priceFor({ tier: ... })`. Vertex's batch tier
+ *    is ~50% of standard; that's not enforced — declare the actual
+ *    published rates.
+ * 4. Bump the package version (semver minor for additions).
+ */
+export const PRICING_TABLE = [
+    // ---------------------------------------------------------------------
+    // Vertex Gemini 3.x (200K-context breakpoint applies to 3.1 Pro)
+    // ---------------------------------------------------------------------
+    // gemini-3.1-pro-preview standard interactive
+    // ≤200K: $2/M in, $12/M out  ;  >200K: $4/M in, $18/M out
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3.1-pro',
+        tier: 'standard',
+        inputPer1M: 2.0,
+        outputPer1M: 12.0,
+        contextTierBreakpoint: 200_000,
+        contextTierAbove: { inputPer1M: 4.0, outputPer1M: 18.0 },
+    },
+    // gemini-3.1-pro-preview flex/batch (sb-srnl 2026-05-07 verified)
+    // ≤200K: $1/M in, $6/M out  ;  >200K: $2/M in, $9/M out
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3.1-pro',
+        tier: 'batch',
+        inputPer1M: 1.0,
+        outputPer1M: 6.0,
+        contextTierBreakpoint: 200_000,
+        contextTierAbove: { inputPer1M: 2.0, outputPer1M: 9.0 },
+    },
+    // gemini-3.1-flash-lite-preview standard
+    // Flat: $0.25/M in, $1.50/M out (no breakpoint per public table)
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3.1-flash-lite',
+        tier: 'standard',
+        inputPer1M: 0.25,
+        outputPer1M: 1.5,
+    },
+    // gemini-3.1-flash-lite-preview flex/batch
+    // Flat: $0.13/M in, $0.75/M out
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3.1-flash-lite',
+        tier: 'batch',
+        inputPer1M: 0.13,
+        outputPer1M: 0.75,
+    },
+    // gemini-3-pro-preview — placeholder pricing using 3.1 sibling rates
+    // (Vertex hasn't published separate 3.0 list prices as of 2026-05-07)
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3-pro',
+        tier: 'standard',
+        inputPer1M: 2.0,
+        outputPer1M: 12.0,
+        contextTierBreakpoint: 200_000,
+        contextTierAbove: { inputPer1M: 4.0, outputPer1M: 18.0 },
+    },
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3-pro',
+        tier: 'batch',
+        inputPer1M: 1.0,
+        outputPer1M: 6.0,
+        contextTierBreakpoint: 200_000,
+        contextTierAbove: { inputPer1M: 2.0, outputPer1M: 9.0 },
+    },
+    // gemini-3-flash-preview — placeholder using 3.1 flash-lite sibling rates
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3-flash',
+        tier: 'standard',
+        inputPer1M: 0.25,
+        outputPer1M: 1.5,
+    },
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-3-flash',
+        tier: 'batch',
+        inputPer1M: 0.13,
+        outputPer1M: 0.75,
+    },
+    // ---------------------------------------------------------------------
+    // Vertex Gemini 2.5 (no context-tier breakpoint per public table)
+    // ---------------------------------------------------------------------
+    // gemini-2.5-pro standard: $1.25/M in, $10/M out
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-2.5-pro',
+        tier: 'standard',
+        inputPer1M: 1.25,
+        outputPer1M: 10.0,
+    },
+    // gemini-2.5-pro batch: ~50% of standard
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-2.5-pro',
+        tier: 'batch',
+        inputPer1M: 0.625,
+        outputPer1M: 5.0,
+    },
+    // gemini-2.5-flash standard: $0.075/M in, $0.30/M out
+    // (startup-builder/llm-vertex-batch source — supersedes icps's stale rate)
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-2.5-flash',
+        tier: 'standard',
+        inputPer1M: 0.075,
+        outputPer1M: 0.3,
+    },
+    // gemini-2.5-flash batch: ~50% of standard
+    {
+        provider: 'vertex',
+        slug: 'vertex/gemini-2.5-flash',
+        tier: 'batch',
+        inputPer1M: 0.0375,
+        outputPer1M: 0.15,
+    },
+    // ---------------------------------------------------------------------
+    // Bedrock Anthropic Claude (flat pricing — no context-tier breakpoint)
+    // ---------------------------------------------------------------------
+    // claude-opus-4-7: $15/M in, $75/M out
+    {
+        provider: 'bedrock',
+        slug: 'bedrock/claude-opus-4-7',
+        tier: 'standard',
+        inputPer1M: 15.0,
+        outputPer1M: 75.0,
+    },
+    // claude-opus-4-6: same as 4-7
+    {
+        provider: 'bedrock',
+        slug: 'bedrock/claude-opus-4-6',
+        tier: 'standard',
+        inputPer1M: 15.0,
+        outputPer1M: 75.0,
+    },
+    // claude-sonnet-4-7: $3/M in, $15/M out
+    {
+        provider: 'bedrock',
+        slug: 'bedrock/claude-sonnet-4-7',
+        tier: 'standard',
+        inputPer1M: 3.0,
+        outputPer1M: 15.0,
+    },
+    // claude-sonnet-4-6: same as 4-7
+    {
+        provider: 'bedrock',
+        slug: 'bedrock/claude-sonnet-4-6',
+        tier: 'standard',
+        inputPer1M: 3.0,
+        outputPer1M: 15.0,
+    },
+    // claude-haiku-4-5: $1/M in, $5/M out
+    {
+        provider: 'bedrock',
+        slug: 'bedrock/claude-haiku-4-5',
+        tier: 'standard',
+        inputPer1M: 1.0,
+        outputPer1M: 5.0,
+    },
+    // ---------------------------------------------------------------------
+    // Google AI Studio embeddings
+    // ---------------------------------------------------------------------
+    // gemini-embedding-2: $0.15/M input (no output side — pure embedding SKU)
+    {
+        provider: 'google-ai-studio',
+        slug: 'aistudio/gemini-embedding-2',
+        tier: 'standard',
+        inputPer1M: 0.15,
+        outputPer1M: 0,
+    },
+];
+//# sourceMappingURL=table.js.map

package/dist/pricing/table.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"table.js","sourceRoot":"","sources":["../../src/pricing/table.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA0CG;AAIH,MAAM,CAAC,MAAM,aAAa,GAA4B;IACpD,wEAAwE;IACxE,iEAAiE;IACjE,wEAAwE;IAExE,8CAA8C;IAC9C,0DAA0D;IAC1D;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,IAAI;QACjB,qBAAqB,EAAE,OAAO;QAC9B,gBAAgB,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,IAAI,EAAE;KACzD;IAED,kEAAkE;IAClE,wDAAwD;IACxD;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,GAAG;QAChB,qBAAqB,EAAE,OAAO;QAC9B,gBAAgB,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,GAAG,EAAE;KACxD;IAED,yCAAyC;IACzC,iEAAiE;IACjE;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,8BAA8B;QACpC,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,GAAG;KACjB;IAED,2CAA2C;IAC3C,gCAAgC;IAChC;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,8BAA8B;QACpC,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,IAAI;KAClB;IAED,qEAAqE;IACrE,sEAAsE;IACtE;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,qBAAqB;QAC3B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,IAAI;QACjB,qBAAqB,EAAE,OAAO;QAC9B,gBAAgB,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,IAAI,EAAE;KACzD;IACD;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,qBAAqB;QAC3B,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,GAAG;QAChB,qBAAqB,EAAE,OAAO;QAC9B,gBAAgB,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,GAAG,EAAE;KACxD;IAED,0EAA0E;IAC1E;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,GAAG;KACjB;IACD;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,IAAI;KAClB;IAED,wEAAwE;IACxE,kEAAkE;IAClE,wEAAwE;IAExE,iDAAiD;IACjD;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,IAAI;KAClB;IAED,yCAAyC;IACzC;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,uBAAuB;QAC7B,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,KAAK;QACjB,WAAW,EAAE,GAAG;KACjB;IAED,sDAAsD;IACtD,2EAA2E;IAC3E;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,yBAAyB;QAC/B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,KAAK;QACjB,WAAW,EAAE,GAAG;KACjB;IAED,2CAA2C;IAC3C;QACE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE,yBAAyB;QAC/B,IAAI,EAAE,OAAO;QACb,UAAU,EAAE,MAAM;QAClB,WAAW,EAAE,IAAI;KAClB;IAED,wEAAwE;IACxE,uEAAuE;IACvE,wEAAwE;IAExE,uCAAuC;IACvC;QACE,QAAQ,EAAE,SAAS;QACnB,IAAI,EAAE,yBAAyB;QAC/B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,IAAI;KAClB;IAED,+BAA+B;IAC/B;QACE,QAAQ,EAAE,SAAS;QACnB,IAAI,EAAE,yBAAyB;QAC/B,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,IAAI;KAClB;IAED,wCAAwC;IACxC;QACE,QAAQ,EAAE,SAAS;QACnB,IAAI,EAAE,2BAA2B;QACjC,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,IAAI;KAClB;IAED,iCAAiC;IACjC;QACE,QAAQ,EAAE,SAAS;QACnB,IAAI,EAAE,2BAA2B;QACjC,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,IAAI;KAClB;IAED,sCAAsC;IACtC;QACE,QAAQ,EAAE,SAAS;QACnB,IAAI,EAAE,0BAA0B;QAChC,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,GAAG;KACjB;IAED,wEAAwE;IACxE,8BAA8B;IAC9B,wEAAwE;IAExE,0EAA0E;IAC1E;QACE,QAAQ,EAAE,kBAAkB;QAC5B,IAAI,EAAE,6BAA6B;QACnC,IAAI,EAAE,UAAU;QAChB,UAAU,EAAE,IAAI;QAChB,WAAW,EAAE,CAAC;KACf;CACF,CAAA"}

package/dist/pricing/types.d.ts

@@ -0,0 +1,84 @@
+/**
+ * language-models / pricing — type definitions for the canonical pricing table.
+ *
+ * Schema design notes (sb-ncer 2026-05-07):
+ *
+ * 1. **Provider** is the upstream API surface, not the model family. AWS
+ *    Bedrock hosts Anthropic Claude, but `provider: 'bedrock'` (not
+ *    `'anthropic'`) — same model, different cost when consumed via
+ *    Anthropic's first-party API key (provider: 'anthropic' would be a
+ *    DIFFERENT row with potentially different rates).
+ *
+ * 2. **Slug** is the caller-facing string: `<provider>/<short-name>`.
+ *    Multiple slugs (the cascade short slug, the SDK-native id, etc.)
+ *    can map to the same logical SKU — but for the canonical primitive
+ *    we only carry the cascade short slug (`vertex/gemini-3.1-pro`,
+ *    `bedrock/claude-opus-4-7`). Adapter packages can layer their own
+ *    rewrite tables on top.
+ *
+ * 3. **Tier** distinguishes pricing modes for the same SKU:
+ *    - `standard`: synchronous interactive pricing (full price)
+ *    - `batch`: async batch-prediction pricing (typically 50% discount)
+ *    - `flex`: flex-tier pricing (Vertex's name for batch, kept as alias)
+ *    - `provisioned`: provisioned-throughput pricing (per-hour, not
+ *      currently modeled — placeholder for future PT entries)
+ *
+ * 4. **contextTierBreakpoint + contextTierAbove**: Gemini 3.x SKUs apply
+ *    a 2× rate above 200K input tokens; Anthropic Claude pricing is flat.
+ *    Optional fields — when absent, the base rate applies regardless of
+ *    input size.
+ */
+export type Provider = 'vertex' | 'bedrock' | 'openai' | 'anthropic' | 'google-ai-studio';
+export type PricingTier = 'standard' | 'batch' | 'flex' | 'provisioned';
+/** Rates expressed in USD per 1,000,000 tokens. */
+export interface RateBlock {
+    /** USD per 1M input tokens. */
+    readonly inputPer1M: number;
+    /** USD per 1M output (completion) tokens. */
+    readonly outputPer1M: number;
+    /**
+     * Optional USD per 1M cached input tokens (prompt-caching tier). When
+     * absent, callers should fall back to inputPer1M (no cache discount).
+     */
+    readonly cachedInputPer1M?: number;
+}
+/**
+ * Single canonical pricing row. Identity is `(slug, tier)` — provider is
+ * derived from the slug prefix and stored explicitly only for tooling
+ * convenience.
+ */
+export interface ModelPricing extends RateBlock {
+    readonly provider: Provider;
+    readonly slug: string;
+    readonly tier: PricingTier;
+    /**
+     * Token count at which pricing changes (inclusive — i.e. inputs >=
+     * breakpoint use contextTierAbove). Currently only Gemini 3.x SKUs
+     * have a breakpoint (200_000). Anthropic Claude has flat pricing.
+     */
+    readonly contextTierBreakpoint?: number;
+    /** Rates that apply when inputTokens >= contextTierBreakpoint. */
+    readonly contextTierAbove?: RateBlock;
+}
+export interface PriceForArgs {
+    readonly slug: string;
+    readonly tier: PricingTier;
+    readonly inputTokens: number;
+    readonly outputTokens: number;
+    /**
+     * Optional cached input tokens (subset of inputTokens that hit a
+     * prompt-caching tier). Billed at cachedInputPer1M when the row
+     * defines it; otherwise at inputPer1M (i.e. no discount).
+     */
+    readonly cachedInputTokens?: number;
+}
+export interface PriceForResult {
+    readonly inputUsd: number;
+    readonly outputUsd: number;
+    readonly totalUsd: number;
+}
+export interface HasPricingArgs {
+    readonly slug: string;
+    readonly tier: PricingTier;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/pricing/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/pricing/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA6BG;AAEH,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,QAAQ,GAAG,WAAW,GAAG,kBAAkB,CAAA;AAEzF,MAAM,MAAM,WAAW,GAAG,UAAU,GAAG,OAAO,GAAG,MAAM,GAAG,aAAa,CAAA;AAEvE,mDAAmD;AACnD,MAAM,WAAW,SAAS;IACxB,+BAA+B;IAC/B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAA;IAC3B,6CAA6C;IAC7C,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B;;;OAGG;IACH,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;CACnC;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAa,SAAQ,SAAS;IAC7C,QAAQ,CAAC,QAAQ,EAAE,QAAQ,CAAA;IAC3B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;IAC1B;;;;OAIG;IACH,QAAQ,CAAC,qBAAqB,CAAC,EAAE,MAAM,CAAA;IACvC,kEAAkE;IAClE,QAAQ,CAAC,gBAAgB,CAAC,EAAE,SAAS,CAAA;CACtC;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;IAC1B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;IAC7B;;;;OAIG;IACH,QAAQ,CAAC,iBAAiB,CAAC,EAAE,MAAM,CAAA;CACpC;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;CAC3B"}

package/dist/pricing/types.js

@@ -0,0 +1,32 @@
+/**
+ * language-models / pricing — type definitions for the canonical pricing table.
+ *
+ * Schema design notes (sb-ncer 2026-05-07):
+ *
+ * 1. **Provider** is the upstream API surface, not the model family. AWS
+ *    Bedrock hosts Anthropic Claude, but `provider: 'bedrock'` (not
+ *    `'anthropic'`) — same model, different cost when consumed via
+ *    Anthropic's first-party API key (provider: 'anthropic' would be a
+ *    DIFFERENT row with potentially different rates).
+ *
+ * 2. **Slug** is the caller-facing string: `<provider>/<short-name>`.
+ *    Multiple slugs (the cascade short slug, the SDK-native id, etc.)
+ *    can map to the same logical SKU — but for the canonical primitive
+ *    we only carry the cascade short slug (`vertex/gemini-3.1-pro`,
+ *    `bedrock/claude-opus-4-7`). Adapter packages can layer their own
+ *    rewrite tables on top.
+ *
+ * 3. **Tier** distinguishes pricing modes for the same SKU:
+ *    - `standard`: synchronous interactive pricing (full price)
+ *    - `batch`: async batch-prediction pricing (typically 50% discount)
+ *    - `flex`: flex-tier pricing (Vertex's name for batch, kept as alias)
+ *    - `provisioned`: provisioned-throughput pricing (per-hour, not
+ *      currently modeled — placeholder for future PT entries)
+ *
+ * 4. **contextTierBreakpoint + contextTierAbove**: Gemini 3.x SKUs apply
+ *    a 2× rate above 200K input tokens; Anthropic Claude pricing is flat.
+ *    Optional fields — when absent, the base rate applies regardless of
+ *    input size.
+ */
+export {};
+//# sourceMappingURL=types.js.map

package/dist/pricing/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/pricing/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA6BG"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "language-models",
-  "version": "2.1.3",
+  "version": "2.3.0",
   "description": "Model listing and resolution for LLM providers",
   "type": "module",
   "main": "dist/index.js",
@@ -9,8 +9,21 @@
     ".": {
       "import": "./dist/index.js",
       "types": "./dist/index.d.ts"
+    },
+    "./pricing": {
+      "import": "./dist/pricing/index.js",
+      "types": "./dist/pricing/index.d.ts"
     }
   },
+  "scripts": {
+    "build": "tsc -p tsconfig.json && cp -r data dist/",
+    "dev": "tsc -p tsconfig.json --watch",
+    "test": "vitest run",
+    "typecheck": "tsc --noEmit",
+    "lint": "eslint .",
+    "clean": "rm -rf dist",
+    "fetch-models": "npx tsx scripts/fetch-models.ts"
+  },
   "keywords": [
     "ai",
     "llm",
@@ -19,14 +32,5 @@
     "primitives"
   ],
   "license": "MIT",
-  "dependencies": {},
-  "scripts": {
-    "build": "tsc -p tsconfig.json && cp -r data dist/",
-    "dev": "tsc -p tsconfig.json --watch",
-    "test": "vitest",
-    "typecheck": "tsc --noEmit",
-    "lint": "eslint .",
-    "clean": "rm -rf dist",
-    "fetch-models": "npx tsx scripts/fetch-models.ts"
-  }
-}
+  "dependencies": {}
+}

package/src/index.ts

@@ -16,6 +16,47 @@ export {
   type ModelInfo,
   type ProviderEndpoint,
   type ResolvedModel,
-  type DirectProvider
+  type DirectProvider,
 } from './models.js'
 export { ALIASES } from './aliases.js'
+
+// Per-model resilience and tier policy data.
+// The runtime *machinery* (RetryPolicy, CircuitBreaker, FallbackChain) lives
+// in `ai-functions`; the *data* (which categories are retryable, which models
+// to fall back to, which batch tiers are eligible) lives here alongside the
+// catalog because it's a per-model concern.
+export {
+  policyFor,
+  derivePolicy,
+  defaultPolicy,
+  resetPolicyCache,
+  listAliases,
+  DEFAULT_RETRY,
+  DEFAULT_CIRCUIT_BREAKER,
+  type ModelPolicy,
+  type RetryPolicyData,
+  type CircuitBreakerPolicyData,
+  type BatchTier,
+  type ErrorCategoryName,
+  type ErrorMapping,
+} from './policy.js'
+
+// =============================================================================
+// Pricing (consolidated from former @primitives/llm-pricing — see
+// `./pricing/` for the source. Also re-exported under the subpath
+// `language-models/pricing` for surgical imports.)
+// =============================================================================
+export {
+  PRICING_TABLE,
+  priceFor,
+  listSlugs,
+  hasPricing,
+  rowsForSlug,
+  type ModelPricing,
+  type PricingTier,
+  type Provider,
+  type RateBlock,
+  type PriceForArgs,
+  type PriceForResult,
+  type HasPricingArgs,
+} from './pricing/index.js'

package/src/models.ts

@@ -63,7 +63,7 @@ export function list(): ModelInfo[] {
  * Get a model by exact ID
  */
 export function get(id: string): ModelInfo | undefined {
-  return loadModels().find(m => m.id === id)
+  return loadModels().find((m) => m.id === id)
 }
 
 /**
@@ -72,9 +72,8 @@ export function get(id: string): ModelInfo | undefined {
  */
 export function search(query: string): ModelInfo[] {
   const q = query.toLowerCase()
-  return loadModels().filter(m =>
-    m.id.toLowerCase().includes(q) ||
-    m.name.toLowerCase().includes(q)
+  return loadModels().filter(
+    (m) => m.id.toLowerCase().includes(q) || m.name.toLowerCase().includes(q)
   )
 }
 
@@ -123,7 +122,7 @@ export function resolve(input: string): string {
  * These providers have special capabilities like MCP, extended thinking, etc.
  */
 export const DIRECT_PROVIDERS = ['openai', 'anthropic', 'google'] as const
-export type DirectProvider = typeof DIRECT_PROVIDERS[number]
+export type DirectProvider = (typeof DIRECT_PROVIDERS)[number]
 
 /**
  * Result of resolving a model with provider routing info
@@ -164,11 +163,8 @@ export function resolveWithProvider(input: string): ResolvedModel {
 
   const supportsDirectRouting = (DIRECT_PROVIDERS as readonly string[]).includes(provider)
 
-  return {
-    id,
-    provider,
-    providerModelId: model?.provider_model_id,
-    supportsDirectRouting,
-    model
-  }
+  const result: ResolvedModel = { id, provider, supportsDirectRouting }
+  if (model) result.model = model
+  if (model?.provider_model_id) result.providerModelId = model.provider_model_id
+  return result
 }