@newrelic/preflight 1.63.2 → 1.63.3

package/dist/shared/pricing-data.js

@@ -16,10 +16,14 @@ export const MODEL_ALIASES = {
     'claude-opus-4': 'claude-opus-4-8',
     'claude-sonnet-4': 'claude-sonnet-4-6',
     'claude-haiku-4': 'claude-haiku-4-5',
+    'claude-haiku-5': 'claude-haiku-5-5',
     'claude-haiku-3-5': 'claude-haiku-3-5-20241022',
     // Google Gemini current-gen shortcuts
     'gemini-3.5': 'gemini-3.5-flash',
     'gemini-3.1-pro': 'gemini-3.1-pro-preview',
+    // Same model and price as 3.1 Pro Preview, with a tool-use-tuned endpoint
+    // (ai.google.dev lists both IDs under one price).
+    'gemini-3.1-pro-preview-customtools': 'gemini-3.1-pro-preview',
     // No alias for 'gemini-3.1-flash-lite' — it's now an exact-match table key
     // in its own right (GA'd, dropped the "-preview" suffix). An earlier
     // revision of this file aliased it to 'gemini-3.1-flash-lite-preview',
@@ -38,16 +42,22 @@ export const MODEL_ALIASES = {
     // this file aliased 'gpt-5' -> 'gpt-5.5', which over-billed real `gpt-5`
     // calls by several times ($1.25/$10 actual vs $5/$30 aliased).
     'gpt-5.6': 'gpt-5.6-sol',
+    // OpenAI's bare 'gpt-4' API alias points at the gpt-4-0613 snapshot. Pinned
+    // so it does not depend on forward-prefix matching.
+    'gpt-4': 'gpt-4-0613',
+    // Mistral: bare family name routes to the current generation, not to
+    // whichever 'mistral-large-*' key forward-prefix matching happens to pick.
+    'mistral-large': 'mistral-large-4',
     // AWS Bedrock geo cross-region inference profile IDs. These are a
     // different kind of alias than the family-name shortcuts above — each is
     // a real, literal AWS inference-profile ID, routed here because AWS bills
     // every region below identically to the bare `anthropic.*` "in-region"
     // key it points at (see the Bedrock section's tier comment in
-    // DEFAULT_PRICING_TABLE for the pricing rationale). Not verified that
-    // every region is an actual valid profile for every model — AWS's
-    // regional rollout varies per model, but an alias for a region AWS
-    // doesn't actually offer is simply unreachable in practice, not a source
-    // of mispricing.
+    // DEFAULT_PRICING_TABLE for the pricing rationale). The in./apac. and
+    // non-Claude entries further down are taken from AWS model cards. The
+    // us./eu./au./jp. Claude entries are not verified per model — AWS's
+    // regional rollout varies, but an alias for a region AWS doesn't actually
+    // offer is simply unreachable in practice, not a source of mispricing.
     'us.anthropic.claude-sonnet-5': 'anthropic.claude-sonnet-5',
     'eu.anthropic.claude-sonnet-5': 'anthropic.claude-sonnet-5',
     'au.anthropic.claude-sonnet-5': 'anthropic.claude-sonnet-5',
@@ -76,6 +86,106 @@ export const MODEL_ALIASES = {
     'eu.anthropic.claude-haiku-4-5-20251001-v1:0': 'anthropic.claude-haiku-4-5-20251001-v1:0',
     'au.anthropic.claude-haiku-4-5-20251001-v1:0': 'anthropic.claude-haiku-4-5-20251001-v1:0',
     'jp.anthropic.claude-haiku-4-5-20251001-v1:0': 'anthropic.claude-haiku-4-5-20251001-v1:0',
+    'us.anthropic.claude-sonnet-5-5': 'anthropic.claude-sonnet-5-5',
+    'eu.anthropic.claude-sonnet-5-5': 'anthropic.claude-sonnet-5-5',
+    'au.anthropic.claude-sonnet-5-5': 'anthropic.claude-sonnet-5-5',
+    'jp.anthropic.claude-sonnet-5-5': 'anthropic.claude-sonnet-5-5',
+    'us.anthropic.claude-opus-5-5': 'anthropic.claude-opus-5-5',
+    'eu.anthropic.claude-opus-5-5': 'anthropic.claude-opus-5-5',
+    'au.anthropic.claude-opus-5-5': 'anthropic.claude-opus-5-5',
+    'jp.anthropic.claude-opus-5-5': 'anthropic.claude-opus-5-5',
+    'us.anthropic.claude-fable-5-1': 'anthropic.claude-fable-5-1',
+    'eu.anthropic.claude-fable-5-1': 'anthropic.claude-fable-5-1',
+    'au.anthropic.claude-fable-5-1': 'anthropic.claude-fable-5-1',
+    'jp.anthropic.claude-fable-5-1': 'anthropic.claude-fable-5-1',
+    'us.anthropic.claude-mythos-5': 'anthropic.claude-mythos-5',
+    'eu.anthropic.claude-mythos-5': 'anthropic.claude-mythos-5',
+    'au.anthropic.claude-mythos-5': 'anthropic.claude-mythos-5',
+    'jp.anthropic.claude-mythos-5': 'anthropic.claude-mythos-5',
+    'us.anthropic.claude-mythos-5-1': 'anthropic.claude-mythos-5-1',
+    'eu.anthropic.claude-mythos-5-1': 'anthropic.claude-mythos-5-1',
+    'au.anthropic.claude-mythos-5-1': 'anthropic.claude-mythos-5-1',
+    'jp.anthropic.claude-mythos-5-1': 'anthropic.claude-mythos-5-1',
+    'us.anthropic.claude-mythos-preview': 'anthropic.claude-mythos-preview',
+    'eu.anthropic.claude-mythos-preview': 'anthropic.claude-mythos-preview',
+    'au.anthropic.claude-mythos-preview': 'anthropic.claude-mythos-preview',
+    'jp.anthropic.claude-mythos-preview': 'anthropic.claude-mythos-preview',
+    'us.anthropic.claude-opus-4-6-v1': 'anthropic.claude-opus-4-6-v1',
+    'eu.anthropic.claude-opus-4-6-v1': 'anthropic.claude-opus-4-6-v1',
+    'au.anthropic.claude-opus-4-6-v1': 'anthropic.claude-opus-4-6-v1',
+    'jp.anthropic.claude-opus-4-6-v1': 'anthropic.claude-opus-4-6-v1',
+    'us.anthropic.claude-opus-4-5-20251101-v1:0': 'anthropic.claude-opus-4-5-20251101-v1:0',
+    'eu.anthropic.claude-opus-4-5-20251101-v1:0': 'anthropic.claude-opus-4-5-20251101-v1:0',
+    'au.anthropic.claude-opus-4-5-20251101-v1:0': 'anthropic.claude-opus-4-5-20251101-v1:0',
+    'jp.anthropic.claude-opus-4-5-20251101-v1:0': 'anthropic.claude-opus-4-5-20251101-v1:0',
+    'us.anthropic.claude-sonnet-4-5-20250929-v1:0': 'anthropic.claude-sonnet-4-5-20250929-v1:0',
+    'eu.anthropic.claude-sonnet-4-5-20250929-v1:0': 'anthropic.claude-sonnet-4-5-20250929-v1:0',
+    'au.anthropic.claude-sonnet-4-5-20250929-v1:0': 'anthropic.claude-sonnet-4-5-20250929-v1:0',
+    'jp.anthropic.claude-sonnet-4-5-20250929-v1:0': 'anthropic.claude-sonnet-4-5-20250929-v1:0',
+    'us.anthropic.claude-sonnet-4-20250514-v1:0': 'anthropic.claude-sonnet-4-20250514-v1:0',
+    'eu.anthropic.claude-sonnet-4-20250514-v1:0': 'anthropic.claude-sonnet-4-20250514-v1:0',
+    'au.anthropic.claude-sonnet-4-20250514-v1:0': 'anthropic.claude-sonnet-4-20250514-v1:0',
+    'jp.anthropic.claude-sonnet-4-20250514-v1:0': 'anthropic.claude-sonnet-4-20250514-v1:0',
+    'us.anthropic.claude-opus-4-1-20250805-v1:0': 'anthropic.claude-opus-4-1-20250805-v1:0',
+    'eu.anthropic.claude-opus-4-1-20250805-v1:0': 'anthropic.claude-opus-4-1-20250805-v1:0',
+    'au.anthropic.claude-opus-4-1-20250805-v1:0': 'anthropic.claude-opus-4-1-20250805-v1:0',
+    'jp.anthropic.claude-opus-4-1-20250805-v1:0': 'anthropic.claude-opus-4-1-20250805-v1:0',
+    'us.anthropic.claude-opus-4-20250514-v1:0': 'anthropic.claude-opus-4-20250514-v1:0',
+    'eu.anthropic.claude-opus-4-20250514-v1:0': 'anthropic.claude-opus-4-20250514-v1:0',
+    'au.anthropic.claude-opus-4-20250514-v1:0': 'anthropic.claude-opus-4-20250514-v1:0',
+    'jp.anthropic.claude-opus-4-20250514-v1:0': 'anthropic.claude-opus-4-20250514-v1:0',
+    'us.anthropic.claude-3-7-sonnet-20250219-v1:0': 'anthropic.claude-3-7-sonnet-20250219-v1:0',
+    'eu.anthropic.claude-3-7-sonnet-20250219-v1:0': 'anthropic.claude-3-7-sonnet-20250219-v1:0',
+    'au.anthropic.claude-3-7-sonnet-20250219-v1:0': 'anthropic.claude-3-7-sonnet-20250219-v1:0',
+    'jp.anthropic.claude-3-7-sonnet-20250219-v1:0': 'anthropic.claude-3-7-sonnet-20250219-v1:0',
+    'us.anthropic.claude-3-haiku-20240307-v1:0': 'anthropic.claude-3-haiku-20240307-v1:0',
+    'eu.anthropic.claude-3-haiku-20240307-v1:0': 'anthropic.claude-3-haiku-20240307-v1:0',
+    'au.anthropic.claude-3-haiku-20240307-v1:0': 'anthropic.claude-3-haiku-20240307-v1:0',
+    'jp.anthropic.claude-3-haiku-20240307-v1:0': 'anthropic.claude-3-haiku-20240307-v1:0',
+    'us.anthropic.claude-3-sonnet-20240229-v1:0': 'anthropic.claude-3-sonnet-20240229-v1:0',
+    'eu.anthropic.claude-3-sonnet-20240229-v1:0': 'anthropic.claude-3-sonnet-20240229-v1:0',
+    'au.anthropic.claude-3-sonnet-20240229-v1:0': 'anthropic.claude-3-sonnet-20240229-v1:0',
+    'jp.anthropic.claude-3-sonnet-20240229-v1:0': 'anthropic.claude-3-sonnet-20240229-v1:0',
+    // Legacy dated IDs: without these, `us.`/`eu.`-prefixed profiles resolve to
+    // null ($0) because the prefix-match fallback can't strip a region prefix.
+    'us.anthropic.claude-3-5-sonnet-20241022-v2:0': 'anthropic.claude-3-5-sonnet-20241022-v2:0',
+    'eu.anthropic.claude-3-5-sonnet-20241022-v2:0': 'anthropic.claude-3-5-sonnet-20241022-v2:0',
+    'us.anthropic.claude-3-5-haiku-20241022-v1:0': 'anthropic.claude-3-5-haiku-20241022-v1:0',
+    'eu.anthropic.claude-3-5-haiku-20241022-v1:0': 'anthropic.claude-3-5-haiku-20241022-v1:0',
+    'us.anthropic.claude-3-opus-20240229-v1:0': 'anthropic.claude-3-opus-20240229-v1:0',
+    'eu.anthropic.claude-3-opus-20240229-v1:0': 'anthropic.claude-3-opus-20240229-v1:0',
+    // Geo inference profiles for the remaining Bedrock models, per their AWS model cards.
+    'in.anthropic.claude-haiku-4-5-20251001-v1:0': 'anthropic.claude-haiku-4-5-20251001-v1:0',
+    'in.anthropic.claude-opus-5': 'anthropic.claude-opus-5',
+    'in.anthropic.claude-sonnet-5': 'anthropic.claude-sonnet-5',
+    'apac.anthropic.claude-sonnet-4-20250514-v1:0': 'anthropic.claude-sonnet-4-20250514-v1:0',
+    'us.anthropic.claude-haiku-5-5': 'anthropic.claude-haiku-5-5',
+    'eu.anthropic.claude-haiku-5-5': 'anthropic.claude-haiku-5-5',
+    'au.anthropic.claude-haiku-5-5': 'anthropic.claude-haiku-5-5',
+    'jp.anthropic.claude-haiku-5-5': 'anthropic.claude-haiku-5-5',
+    'us.amazon.nova-2-lite-v1:0': 'amazon.nova-2-lite-v1:0',
+    'eu.amazon.nova-2-lite-v1:0': 'amazon.nova-2-lite-v1:0',
+    'jp.amazon.nova-2-lite-v1:0': 'amazon.nova-2-lite-v1:0',
+    'us.amazon.nova-lite-v1:0': 'amazon.nova-lite-v1:0',
+    'eu.amazon.nova-lite-v1:0': 'amazon.nova-lite-v1:0',
+    'us.amazon.nova-micro-v1:0': 'amazon.nova-micro-v1:0',
+    'eu.amazon.nova-micro-v1:0': 'amazon.nova-micro-v1:0',
+    'us.amazon.nova-pro-v1:0': 'amazon.nova-pro-v1:0',
+    'eu.amazon.nova-pro-v1:0': 'amazon.nova-pro-v1:0',
+    'apac.amazon.nova-pro-v1:0': 'amazon.nova-pro-v1:0',
+    'us.meta.llama4-maverick-17b-instruct-v1:0': 'meta.llama4-maverick-17b-instruct-v1:0',
+    'us.meta.llama4-scout-17b-instruct-v1:0': 'meta.llama4-scout-17b-instruct-v1:0',
+    'us.meta.llama3-3-70b-instruct-v1:0': 'meta.llama3-3-70b-instruct-v1:0',
+    'us.meta.llama3-2-11b-instruct-v1:0': 'meta.llama3-2-11b-instruct-v1:0',
+    'us.meta.llama3-2-90b-instruct-v1:0': 'meta.llama3-2-90b-instruct-v1:0',
+    'us.meta.llama3-1-8b-instruct-v1:0': 'meta.llama3-1-8b-instruct-v1:0',
+    'us.meta.llama3-1-70b-instruct-v1:0': 'meta.llama3-1-70b-instruct-v1:0',
+    'us.meta.llama3-2-1b-instruct-v1:0': 'meta.llama3-2-1b-instruct-v1:0',
+    'eu.meta.llama3-2-1b-instruct-v1:0': 'meta.llama3-2-1b-instruct-v1:0',
+    'us.meta.llama3-2-3b-instruct-v1:0': 'meta.llama3-2-3b-instruct-v1:0',
+    'eu.meta.llama3-2-3b-instruct-v1:0': 'meta.llama3-2-3b-instruct-v1:0',
+    'us.mistral.pixtral-large-2502-v1:0': 'mistral.pixtral-large-2502-v1:0',
+    'eu.mistral.pixtral-large-2502-v1:0': 'mistral.pixtral-large-2502-v1:0',
     // Mistral's pricing page now shows a "-latest" alias for each Ministral 3
     // size, routing to the dated snapshot IDs that are this table's actual
     // keys (see the Ministral 3 comment in DEFAULT_PRICING_TABLE).
@@ -86,7 +196,11 @@ export const MODEL_ALIASES = {
 // ---------------------------------------------------------------------------
 // Built-in pricing table — USD per million tokens
 //
-// Rates last verified against vendor public pricing pages on 2026-08-27:
+// Rates last fully verified against vendor public pricing pages on
+// 2026-08-27. Later partial passes: Anthropic 5.5, GPT-6, Bedrock, xAI,
+// Moonshot, Mistral Large 4 and the Copilot stale-entry notes on 2026-10-07;
+// OpenAI long-context cache tiers, xAI cache tiers and Moonshot cache writes
+// on 2026-10-08:
 //   - Anthropic   https://www.anthropic.com/pricing, docs.claude.com
 //   - Google      https://ai.google.dev/gemini-api/docs/pricing
 //   - OpenAI      https://platform.openai.com/docs/pricing, /docs/models
@@ -95,7 +209,8 @@ export const MODEL_ALIASES = {
 //   - Bedrock     https://aws.amazon.com/bedrock/pricing/
 //
 // A handful of entries (marked individually below) are instead sourced from
-// GitHub Copilot's own pricing page, verified 2026-08-27:
+// GitHub Copilot's own pricing page, verified 2026-08-27 (delistings re-checked
+// 2026-10-07):
 //   - GitHub Copilot  https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
 // Copilot's per-token pricing is a verified passthrough of each underlying
 // vendor's own direct-API list price, not a Copilot-specific markup —
@@ -160,7 +275,17 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 12.5,
         contextWindow: 1_000_000,
     },
-    // Anthropic's current top-tier model (replaces Opus 4.8 as the default).
+    // Opus 5.5 is the newest Opus: $4/$20 (below Opus 5's $5/$25), with cache
+    // reads at 0.05x input ($0.20) rather than the usual 0.1x. 5m cache write
+    // is the standard 1.25x ($5).
+    'claude-opus-5-5': {
+        inputPerMTok: 4,
+        outputPerMTok: 20,
+        thinkingPerMTok: 20,
+        cacheReadPerMTok: 0.2,
+        cacheCreationPerMTok: 5,
+        contextWindow: 1_000_000,
+    },
     'claude-opus-5': {
         inputPerMTok: 5,
         outputPerMTok: 25,
@@ -189,6 +314,15 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 2.5,
         contextWindow: 1_000_000,
     },
+    // Same rates as claude-sonnet-5.
+    'claude-sonnet-5-5': {
+        inputPerMTok: 2,
+        outputPerMTok: 10,
+        thinkingPerMTok: 10,
+        cacheReadPerMTok: 0.2,
+        cacheCreationPerMTok: 2.5,
+        contextWindow: 1_000_000,
+    },
     'claude-sonnet-4-6': {
         inputPerMTok: 3,
         outputPerMTok: 15,
@@ -197,6 +331,24 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 3.75,
         contextWindow: 1_000_000,
     },
+    // Haiku 5.5 is priced by prompt length, flat: a prompt over 100k tokens bills
+    // the whole request at 5x input/output and 5x cache reads (platform.claude.com
+    // pricing, verified 2026-10-08). The 5m cache-write tier rate ($0.625 above
+    // 100k) is not modeled; cache writes stay at the base $0.125. 1M context per
+    // the model overview. 'claude-haiku-4' deliberately stays on 4.5.
+    'claude-haiku-5-5': {
+        inputPerMTok: 0.1,
+        outputPerMTok: 0.5,
+        thinkingPerMTok: 0.5,
+        cacheReadPerMTok: 0.01,
+        cacheCreationPerMTok: 0.125,
+        contextWindow: 1_000_000,
+        tierThreshold: 100_000,
+        tierInputPerMTok: 0.5,
+        tierOutputPerMTok: 2.5,
+        tierThinkingPerMTok: 2.5,
+        tierCacheReadPerMTok: 0.05,
+    },
     // Dateless current-gen entry — MODEL_ALIASES['claude-haiku-4'] routes here.
     // Matches the opus/sonnet convention: alias → dateless key → dated legacy key.
     // Update this entry when Haiku 4.x rates change; the dated entry below is
@@ -209,6 +361,8 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 1.25,
         contextWindow: 200_000,
     },
+    // Anthropic lists retirement "not sooner than" 2026-10-15; recheck the
+    // deprecations page on or after that date.
     'claude-haiku-4-5-20251001': {
         inputPerMTok: 1,
         outputPerMTok: 5,
@@ -240,6 +394,8 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 6.25,
         contextWindow: 1_000_000,
     },
+    // Deprecated 2026-09-30, retires 2026-11-30 (replacement claude-sonnet-5-5).
+    // Keep for historical-cost backfill after that date.
     'claude-sonnet-4-5': {
         inputPerMTok: 3,
         outputPerMTok: 15,
@@ -261,6 +417,8 @@ export const DEFAULT_PRICING_TABLE = {
     // base. A future 'claude-opus-4-10' resolves via the dated-key + alias path
     // instead (see pricing.test.ts "13b"), never through this entry — no
     // action needed here when that model ships.
+    // Retired (tentative 2026-08-05 per Anthropic's deprecations page). Kept for
+    // historical-cost backfill.
     'claude-opus-4-1': {
         inputPerMTok: 15,
         outputPerMTok: 75,
@@ -269,6 +427,7 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 18.75,
         contextWindow: 200_000,
     },
+    // Retired 2026-06-15. Kept for historical-cost backfill.
     'claude-sonnet-4-20250514': {
         inputPerMTok: 3,
         outputPerMTok: 15,
@@ -277,6 +436,7 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 3.75,
         contextWindow: 200_000,
     },
+    // Retired 2026-06-15. Kept for historical-cost backfill.
     'claude-opus-4-20250514': {
         inputPerMTok: 15,
         outputPerMTok: 75,
@@ -285,6 +445,7 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 18.75,
         contextWindow: 200_000,
     },
+    // Retired 2026-02-19. Kept for historical-cost backfill.
     'claude-haiku-3-5-20241022': {
         inputPerMTok: 0.8,
         outputPerMTok: 4,
@@ -294,58 +455,59 @@ export const DEFAULT_PRICING_TABLE = {
     },
     // ---- Google Gemini (current generation) ----
     // Gemini 3.6 Flash, 3.7 Flash, and 3.8 Flash are newer generations than 3.5
-    // Flash, released after this table's last full pass. All three currently
+    // Flash, are the current Flash generation. All three currently
     // carry identical promotional pricing that steps up on 2027-01-01
     // (confirmed directly on ai.google.dev/gemini-api/docs/pricing on
     // 2026-09-02 for 3.8, 2026-08-14 for 3.6/3.7): input $0.75->$1.50, output
     // (incl. thinking) $3.75->$7.50. Rates below are the CURRENT (pre-step-up)
     // rate — update all three to the post-2027-01-01 rate once that date
-    // passes. (Google also publishes a context-caching rate, $0.075->$0.15,
-    // but it's deliberately omitted here, consistent with every other Gemini
-    // entry in this table: none of them set cacheReadPerMTok even though
-    // extractGeminiTokens() does report a real cachedContentTokenCount. This
-    // is a pre-existing gap across the whole Gemini section, not something
-    // introduced here — worth a follow-up pass if cached Gemini usage becomes
-    // material.)
+    // passes. The context-caching read rate steps up the same day, $0.075->$0.15
+    // (ai.google.dev, verified 2026-10-08). Cache storage is billed per hour
+    // and is not modeled.
     //
     // 3.8 Flash's own model page states its exact input-token limit as
     // 1,048,576, not a round 1,000,000 — contextWindow below is rounded to
-    // 1_000_000 to match every sibling Gemini entry's convention in this file
-    // (all of which round the same way), not a precision loss introduced here.
+    // 1_000_000 to match every sibling Gemini entry's convention in this file.
     'gemini-3.8-flash': {
         inputPerMTok: 0.75,
+        cacheReadPerMTok: 0.075,
         outputPerMTok: 3.75,
         thinkingPerMTok: 3.75,
         contextWindow: 1_000_000,
     },
     'gemini-3.7-flash': {
         inputPerMTok: 0.75,
+        cacheReadPerMTok: 0.075,
         outputPerMTok: 3.75,
         thinkingPerMTok: 3.75,
         contextWindow: 1_000_000,
     },
     'gemini-3.6-flash': {
         inputPerMTok: 0.75,
+        cacheReadPerMTok: 0.075,
         outputPerMTok: 3.75,
         thinkingPerMTok: 3.75,
         contextWindow: 1_000_000,
     },
     'gemini-3.5-flash': {
         inputPerMTok: 1.5,
+        cacheReadPerMTok: 0.15,
         outputPerMTok: 9,
         thinkingPerMTok: 9,
         contextWindow: 1_000_000,
     },
-    // New — Gemini 3.5 generation's cost-efficient tier, released alongside
-    // 3.5 Flash but not previously in this table.
+    // Gemini 3.5 generation's cost-efficient tier, released alongside 3.5 Flash.
     'gemini-3.5-flash-lite': {
         inputPerMTok: 0.3,
+        cacheReadPerMTok: 0.03,
         outputPerMTok: 2.5,
         thinkingPerMTok: 2.5,
         contextWindow: 1_000_000,
     },
     'gemini-3.1-pro-preview': {
         inputPerMTok: 2,
+        cacheReadPerMTok: 0.2,
+        tierCacheReadPerMTok: 0.4,
         outputPerMTok: 12,
         thinkingPerMTok: 12,
         contextWindow: 1_000_000,
@@ -367,25 +529,46 @@ export const DEFAULT_PRICING_TABLE = {
     // set explicitly or reasoning spend is silently billed at $0.
     'gemini-3.1-flash-lite': {
         inputPerMTok: 0.25,
+        audioInputPerMTok: 0.5,
+        cacheReadPerMTok: 0.025,
+        audioCacheReadPerMTok: 0.05,
         outputPerMTok: 1.5,
         thinkingPerMTok: 1.5,
         contextWindow: 1_000_000,
     },
     'gemini-3.1-flash-lite-preview': {
         inputPerMTok: 0.25,
+        audioInputPerMTok: 0.5,
+        cacheReadPerMTok: 0.025,
+        audioCacheReadPerMTok: 0.05,
         outputPerMTok: 1.5,
         thinkingPerMTok: 1.5,
         contextWindow: 1_000_000,
     },
     'gemini-3-flash-preview': {
         inputPerMTok: 0.5,
+        audioInputPerMTok: 1,
+        cacheReadPerMTok: 0.05,
+        audioCacheReadPerMTok: 0.1,
         outputPerMTok: 3,
         thinkingPerMTok: 3,
         contextWindow: 1_000_000,
     },
+    // Gemini Robotics-ER 2 (text-token rates; image/video/audio inputs bill at the
+    // same per-token rate). Output price includes thinking tokens. $1/$5 is
+    // introductory per ai.google.dev, verified 2026-10-08: Google lists
+    // $2/$10 starting 2027-01-01, so bump these rates once that date passes.
+    'gemini-robotics-er-2-preview': {
+        inputPerMTok: 1,
+        outputPerMTok: 5,
+        thinkingPerMTok: 5,
+        contextWindow: 131_072,
+    },
     // ---- Google Gemini 2.5 ----
     'gemini-2.5-pro': {
         inputPerMTok: 1.25,
+        cacheReadPerMTok: 0.125,
+        tierCacheReadPerMTok: 0.25,
         outputPerMTok: 10,
         thinkingPerMTok: 10,
         contextWindow: 1_000_000,
@@ -396,6 +579,9 @@ export const DEFAULT_PRICING_TABLE = {
     },
     'gemini-2.5-flash': {
         inputPerMTok: 0.3,
+        audioInputPerMTok: 1,
+        cacheReadPerMTok: 0.03,
+        audioCacheReadPerMTok: 0.1,
         outputPerMTok: 2.5,
         thinkingPerMTok: 2.5,
         contextWindow: 1_000_000,
@@ -404,6 +590,9 @@ export const DEFAULT_PRICING_TABLE = {
     // above; same reasoning applies here.
     'gemini-2.5-flash-lite': {
         inputPerMTok: 0.1,
+        audioInputPerMTok: 0.3,
+        cacheReadPerMTok: 0.01,
+        audioCacheReadPerMTok: 0.03,
         outputPerMTok: 0.4,
         thinkingPerMTok: 0.4,
         contextWindow: 1_000_000,
@@ -440,14 +629,63 @@ export const DEFAULT_PRICING_TABLE = {
         tierOutputPerMTok: 0.6,
     },
     // ---- OpenAI (current generation) ----
-    // GPT-5.6 (Sol/Terra/Luna) is OpenAI's current recommended flagship
-    // family, replacing unsuffixed GPT-5.5 as the "top tier" choice. Same
-    // flat long-context tiering as GPT-5.5 (2x input/1.5x output above 272k).
+    // GPT-6 (Astra/Sol/Luna) is OpenAI's newest family. Same flat long-context
+    // tiering as GPT-5.5/5.6 (2x input/1.5x output above 272k), plus 2x cached
+    // input above 272k (stated on each model page, verified 2026-10-08;
+    // tierCacheReadPerMTok). Same 1.25x cache-write multiplier as GPT-5.6 (not
+    // modeled — see the GPT-5.6 note below).
+    // gpt-6-sol is superseded by gpt-6.1-sol but remains billable.
+    'gpt-6-astra': {
+        inputPerMTok: 10,
+        outputPerMTok: 50,
+        cacheReadPerMTok: 1,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierCacheReadPerMTok: 2,
+        tierInputPerMTok: 20,
+        tierOutputPerMTok: 75,
+    },
+    'gpt-6.1-sol': {
+        inputPerMTok: 2,
+        outputPerMTok: 10,
+        cacheReadPerMTok: 0.1,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierCacheReadPerMTok: 0.2,
+        tierInputPerMTok: 4,
+        tierOutputPerMTok: 15,
+    },
+    'gpt-6-sol': {
+        inputPerMTok: 2,
+        outputPerMTok: 10,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 4,
+        tierOutputPerMTok: 15,
+    },
+    'gpt-6-luna': {
+        inputPerMTok: 0.1,
+        outputPerMTok: 0.5,
+        cacheReadPerMTok: 0.01,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierCacheReadPerMTok: 0.02,
+        tierInputPerMTok: 0.2,
+        tierOutputPerMTok: 0.75,
+    },
+    // GPT-5.6 (Sol/Terra/Luna) is the prior-generation flagship family,
+    // superseded by GPT-6. Same flat long-context tiering as GPT-5.5 (2x
+    // input/1.5x output above 272k).
     // Cache writes are billed at 1.25x uncached input on this family — not
-    // modeled here: extractOpenAITokens() in tokens.ts hardcodes
-    // cacheCreationTokens to 0 (OpenAI's usage shape has no cache-write
-    // count), so a cacheCreationPerMTok here would be dead data until that
-    // extractor gains a field for it.
+    // modeled here: no cacheCreationPerMTok is set, so any cache-write tokens
+    // that extractOpenAITokens() reports (prompt_tokens_details.cache_write_tokens)
+    // are billed at $0.
+    // The GPT-5.5/5.6 pages (verified 2026-10-08) say the >272k tier is "2x
+    // input and 1.5x output" and are silent on cached input, so no
+    // tierCacheReadPerMTok is set for them (unlike GPT-6, which states 2x
+    // "input and cache rates"). Cached input above 272k bills at the base rate.
     // $4/$20 is promotional pricing per OpenAI's own model page: "a 20%
     // reduction in input pricing and a 33% reduction in output pricing"
     // versus GPT-5.5 ($5/$30), "available at least through November 21,
@@ -508,7 +746,7 @@ export const DEFAULT_PRICING_TABLE = {
         cacheReadPerMTok: 0.175,
         contextWindow: 400_000,
     },
-    // Prior-generation flagship, superseded by gpt-5.6-sol above but still
+    // Prior-generation flagship, superseded by gpt-5.6-sol and GPT-6 but still
     // billable at these rates.
     'gpt-5.5': {
         inputPerMTok: 5,
@@ -547,6 +785,108 @@ export const DEFAULT_PRICING_TABLE = {
         cacheReadPerMTok: 0.02,
         contextWindow: 400_000,
     },
+    // ---- OpenAI (current-generation catalog gaps, added per issue #47) ----
+    // Text-token models only; audio/image/TTS/embedding SKUs are out of scope.
+    // Rates verified against developers.openai.com/api/docs/pricing.md and each
+    // model page on 2026-10-08. The -pro models re-price the full request above
+    // 272k input, same flat-tier mode as gpt-5.5.
+    // gpt-5.5-cyber is deliberately absent: its model page publishes no context window.
+    'gpt-5.5-pro': {
+        inputPerMTok: 30,
+        outputPerMTok: 180,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierInputPerMTok: 60,
+        tierOutputPerMTok: 270,
+    },
+    'gpt-5.4-pro': {
+        inputPerMTok: 30,
+        outputPerMTok: 180,
+        contextWindow: 1_050_000,
+        tierThreshold: 272_000,
+        tierInputPerMTok: 60,
+        tierOutputPerMTok: 270,
+    },
+    'gpt-5.2': {
+        inputPerMTok: 1.75,
+        outputPerMTok: 14,
+        cacheReadPerMTok: 0.175,
+        contextWindow: 400_000,
+    },
+    'gpt-5.2-pro': {
+        inputPerMTok: 21,
+        outputPerMTok: 168,
+        contextWindow: 400_000,
+    },
+    // Shutdown 2027-04-01 per OpenAI; keep for historical-cost backfill after.
+    'gpt-5.1': {
+        inputPerMTok: 1.25,
+        outputPerMTok: 10,
+        cacheReadPerMTok: 0.125,
+        contextWindow: 400_000,
+    },
+    'gpt-5-nano': {
+        inputPerMTok: 0.05,
+        outputPerMTok: 0.4,
+        cacheReadPerMTok: 0.005,
+        contextWindow: 400_000,
+    },
+    'gpt-5-pro': {
+        inputPerMTok: 15,
+        outputPerMTok: 120,
+        contextWindow: 400_000,
+    },
+    'gpt-5.6-cyber': {
+        inputPerMTok: 12.5,
+        outputPerMTok: 75,
+        cacheReadPerMTok: 1.25,
+        contextWindow: 400_000,
+    },
+    'gpt-4.1': {
+        inputPerMTok: 2,
+        outputPerMTok: 8,
+        cacheReadPerMTok: 0.5,
+        contextWindow: 1_047_576,
+    },
+    'gpt-4.1-mini': {
+        inputPerMTok: 0.4,
+        outputPerMTok: 1.6,
+        cacheReadPerMTok: 0.1,
+        contextWindow: 1_047_576,
+    },
+    'gpt-4.1-nano': {
+        inputPerMTok: 0.1,
+        outputPerMTok: 0.4,
+        cacheReadPerMTok: 0.025,
+        contextWindow: 1_047_576,
+    },
+    'o1-pro': {
+        inputPerMTok: 150,
+        outputPerMTok: 600,
+        contextWindow: 200_000,
+    },
+    'o3-pro': {
+        inputPerMTok: 20,
+        outputPerMTok: 80,
+        contextWindow: 200_000,
+    },
+    'gpt-4o-2024-05-13': {
+        inputPerMTok: 5,
+        outputPerMTok: 15,
+        contextWindow: 128_000,
+    },
+    'gpt-4-0613': {
+        inputPerMTok: 30,
+        outputPerMTok: 60,
+        contextWindow: 8_192,
+    },
+    'gpt-3.5-turbo-instruct': {
+        inputPerMTok: 1.5,
+        outputPerMTok: 2,
+        contextWindow: 4_096,
+    },
+    // Deprecated: gpt-4.1-nano, o1-pro, gpt-4o-2024-05-13 and gpt-4-0613 shut down
+    // 2026-10-23; gpt-5-pro and o3-pro 2026-12-11. Kept for historical-cost backfill.
     // ---- OpenAI (legacy) ----
     // gpt-4o/-mini and their dated snapshots below all publish a cached-input
     // rate (~50% off list, notably less generous than the ~90% discount on
@@ -597,15 +937,17 @@ export const DEFAULT_PRICING_TABLE = {
     // o1-mini and o1-preview were fully shut down by OpenAI (2025-10-27 and
     // 2025-07-28 respectively) — requests to these model IDs now fail.
     // Retained only for historical-cost backfill, same convention as
-    // gemini-2.0-flash / command / command-light below.
+    // gemini-2.0-flash (above) / command / command-light (below).
     'o1-mini': {
         inputPerMTok: 1.1,
         outputPerMTok: 4.4,
+        cacheReadPerMTok: 0.55,
         contextWindow: 128_000,
     },
     'o1-preview': {
         inputPerMTok: 15,
         outputPerMTok: 60,
+        cacheReadPerMTok: 7.5,
         contextWindow: 128_000,
     },
     // Repriced down from its April 2025 launch rate ($10/$40) — $2/$8 is the
@@ -629,6 +971,9 @@ export const DEFAULT_PRICING_TABLE = {
         cacheReadPerMTok: 0.275,
         contextWindow: 200_000,
     },
+    // gpt-4-turbo and gpt-3.5-turbo (and o1, o3-mini, o4-mini above) are on
+    // OpenAI's deprecations page with a 2026-10-23 shutdown. Keep for
+    // historical-cost backfill after that date.
     'gpt-4-turbo': {
         inputPerMTok: 10,
         outputPerMTok: 30,
@@ -641,8 +986,8 @@ export const DEFAULT_PRICING_TABLE = {
     },
     // ---- AWS Bedrock (Converse API pricing for on-demand) ----
     // Current Claude generation via Bedrock
-    // Cache pricing for these models IS now published on the Bedrock pricing
-    // page (it wasn't as of this file's last check) — resolved by matching
+    // Cache pricing for these models is published on the Bedrock pricing
+    // page — resolved by matching
     // each model's known direct-API cache-write/-read multipliers (1.25x /
     // 0.1x of input), which line up exactly with AWS's own numbers. AWS
     // publishes BOTH a 5-minute and a 1-hour cache-write rate; cacheCreationPerMTok
@@ -651,9 +996,9 @@ export const DEFAULT_PRICING_TABLE = {
     // $2/$10 is Anthropic's standing Sonnet 5 price on Bedrock too — not
     // introductory. See the direct-API claude-sonnet-5 comment above.
     //
-    // Bare `anthropic.*` keys below (and the `us./eu./au./jp.` regional-prefix
-    // aliases in MODEL_ALIASES) price the "Geo and In-region Cross-region
-    // Inference" tier — AWS bills these two identically. The `global.*` keys
+    // Bare `anthropic.*` keys below (and the regional-prefix
+    // aliases (us./eu./au./jp./in./apac.) in MODEL_ALIASES) price the "Geo and
+    // In-region Cross-region Inference" tier — AWS bills these two identically. The `global.*` keys
     // further down price the separate, ~10%-cheaper "Global Cross-region
     // Inference" tier. On Bedrock a bare `anthropic.*` ID is genuinely the
     // in-region on-demand ID (confirmed real and directly invocable for this
@@ -665,10 +1010,10 @@ export const DEFAULT_PRICING_TABLE = {
     // claude-sonnet-5 via AWS's own per-SKU pricing data ($2.2/$11/cw$2.75/
     // cr$0.22 vs. global's $2/$10/cw$2.5/cr$0.2) and independently corroborated
     // as AWS's documented flat ~10% policy (not model-specific) on the
-    // global-cross-region-inference docs page. The other 6 models' geo rates
-    // below are DERIVED via that same 1.1x multiplier, not independently
-    // re-verified per model — re-confirm against AWS's pricing data in the
-    // next full audit pass if this table's accuracy is ever in question.
+    // global-cross-region-inference docs page. Some pre-5.5
+    // models' geo rates below were originally DERIVED via that same 1.1x
+    // multiplier; entries annotated as read from AWS's pricing page have since
+    // been read directly. Re-confirm the rest in the next full audit pass.
     'anthropic.claude-sonnet-5': {
         inputPerMTok: 2.2,
         outputPerMTok: 11,
@@ -676,6 +1021,23 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 2.75,
         contextWindow: 1_000_000,
     },
+    // Sonnet 5.5 and Opus 5.5 geo rates are read directly from AWS's pricing
+    // page (2026-10-07), not derived from the 1.1x multiplier. Model IDs
+    // confirmed against AWS's model cards (2026-10-07).
+    'anthropic.claude-sonnet-5-5': {
+        inputPerMTok: 2.2,
+        outputPerMTok: 11,
+        cacheReadPerMTok: 0.22,
+        cacheCreationPerMTok: 2.75,
+        contextWindow: 1_000_000,
+    },
+    'anthropic.claude-opus-5-5': {
+        inputPerMTok: 4.4,
+        outputPerMTok: 22,
+        cacheReadPerMTok: 0.22,
+        cacheCreationPerMTok: 5.5,
+        contextWindow: 1_000_000,
+    },
     'anthropic.claude-fable-5': {
         inputPerMTok: 11,
         outputPerMTok: 55,
@@ -683,6 +1045,14 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 13.75,
         contextWindow: 1_000_000,
     },
+    // Geo rate read from AWS's pricing page (2026-10-07).
+    'anthropic.claude-fable-5-1': {
+        inputPerMTok: 11,
+        outputPerMTok: 55,
+        cacheReadPerMTok: 0.275,
+        cacheCreationPerMTok: 13.75,
+        contextWindow: 1_000_000,
+    },
     'anthropic.claude-opus-5': {
         inputPerMTok: 5.5,
         outputPerMTok: 27.5,
@@ -718,7 +1088,108 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 1.375,
         contextWindow: 200_000,
     },
-    // Global Cross-region Inference tier for the same 7 current-gen models
+    // Haiku 5.5: IDs from its AWS model card, rates from the AWS pricing page (2026-10-07).
+    // The page's "Long Context Pricing" table lists 5x input/output/cache-read; it
+    // states no threshold, so the first-party 100k flat threshold is assumed. Cache
+    // writes stay at the base rate, as on the first-party entry.
+    'anthropic.claude-haiku-5-5': {
+        inputPerMTok: 0.11,
+        outputPerMTok: 0.55,
+        thinkingPerMTok: 0.55,
+        cacheReadPerMTok: 0.011,
+        cacheCreationPerMTok: 0.1375,
+        contextWindow: 1_000_000,
+        tierThreshold: 100_000,
+        tierInputPerMTok: 0.55,
+        tierOutputPerMTok: 2.75,
+        tierThinkingPerMTok: 2.75,
+        tierCacheReadPerMTok: 0.055,
+    },
+    // Older and Mythos-line Claude models, geo/in-region rates read from AWS's
+    // pricing page (2026-10-07); IDs confirmed against AWS model cards except
+    // Mythos Preview, Opus 4, 3.7 Sonnet and 3 Sonnet (no card; AWS still prices them).
+    // Mythos 5 is bedrock-mantle (Messages API) only, with no Geo or Global profile.
+    'anthropic.claude-mythos-5': {
+        inputPerMTok: 11,
+        outputPerMTok: 55,
+        cacheReadPerMTok: 1.1,
+        cacheCreationPerMTok: 13.75,
+        contextWindow: 1_000_000,
+    },
+    'anthropic.claude-mythos-5-1': {
+        inputPerMTok: 11,
+        outputPerMTok: 55,
+        cacheReadPerMTok: 0.275,
+        cacheCreationPerMTok: 13.75,
+        contextWindow: 1_000_000,
+    },
+    'anthropic.claude-mythos-preview': {
+        inputPerMTok: 27.5,
+        outputPerMTok: 137.5,
+        cacheReadPerMTok: 2.75,
+        cacheCreationPerMTok: 34.375,
+        contextWindow: 1_000_000,
+    },
+    'anthropic.claude-opus-4-6-v1': {
+        inputPerMTok: 5.5,
+        outputPerMTok: 27.5,
+        cacheReadPerMTok: 0.55,
+        cacheCreationPerMTok: 6.875,
+        contextWindow: 1_000_000,
+    },
+    'anthropic.claude-opus-4-5-20251101-v1:0': {
+        inputPerMTok: 5.5,
+        outputPerMTok: 27.5,
+        cacheReadPerMTok: 0.55,
+        cacheCreationPerMTok: 6.875,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-sonnet-4-5-20250929-v1:0': {
+        inputPerMTok: 3.3,
+        outputPerMTok: 16.5,
+        cacheReadPerMTok: 0.33,
+        cacheCreationPerMTok: 4.125,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-sonnet-4-20250514-v1:0': {
+        inputPerMTok: 3,
+        outputPerMTok: 15,
+        cacheReadPerMTok: 0.3,
+        cacheCreationPerMTok: 3.75,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-opus-4-1-20250805-v1:0': {
+        inputPerMTok: 15,
+        outputPerMTok: 75,
+        cacheReadPerMTok: 1.5,
+        cacheCreationPerMTok: 18.75,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-opus-4-20250514-v1:0': {
+        inputPerMTok: 15,
+        outputPerMTok: 75,
+        cacheReadPerMTok: 1.5,
+        cacheCreationPerMTok: 18.75,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-3-7-sonnet-20250219-v1:0': {
+        inputPerMTok: 3,
+        outputPerMTok: 15,
+        cacheReadPerMTok: 0.3,
+        cacheCreationPerMTok: 3.75,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-3-haiku-20240307-v1:0': {
+        inputPerMTok: 0.25,
+        outputPerMTok: 1.25,
+        contextWindow: 200_000,
+    },
+    'anthropic.claude-3-sonnet-20240229-v1:0': {
+        inputPerMTok: 3,
+        outputPerMTok: 15,
+        contextWindow: 200_000,
+    },
+    // Global Cross-region Inference tier for the Bedrock Claude models
     // above — ~10% cheaper, see the tier comment there. Real, directly
     // invocable Bedrock model IDs (confirmed via AWS's model-card pages), not
     // this table's own naming convention.
@@ -729,6 +1200,20 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 2.5,
         contextWindow: 1_000_000,
     },
+    'global.anthropic.claude-sonnet-5-5': {
+        inputPerMTok: 2,
+        outputPerMTok: 10,
+        cacheReadPerMTok: 0.2,
+        cacheCreationPerMTok: 2.5,
+        contextWindow: 1_000_000,
+    },
+    'global.anthropic.claude-opus-5-5': {
+        inputPerMTok: 4,
+        outputPerMTok: 20,
+        cacheReadPerMTok: 0.2,
+        cacheCreationPerMTok: 5,
+        contextWindow: 1_000_000,
+    },
     'global.anthropic.claude-fable-5': {
         inputPerMTok: 10,
         outputPerMTok: 50,
@@ -771,12 +1256,70 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 1.25,
         contextWindow: 200_000,
     },
+    'global.anthropic.claude-opus-4-6-v1': {
+        inputPerMTok: 5,
+        outputPerMTok: 25,
+        cacheReadPerMTok: 0.5,
+        cacheCreationPerMTok: 6.25,
+        contextWindow: 1_000_000,
+    },
+    'global.anthropic.claude-opus-4-5-20251101-v1:0': {
+        inputPerMTok: 5,
+        outputPerMTok: 25,
+        cacheReadPerMTok: 0.5,
+        cacheCreationPerMTok: 6.25,
+        contextWindow: 200_000,
+    },
+    'global.anthropic.claude-sonnet-4-5-20250929-v1:0': {
+        inputPerMTok: 3,
+        outputPerMTok: 15,
+        cacheReadPerMTok: 0.3,
+        cacheCreationPerMTok: 3.75,
+        contextWindow: 200_000,
+    },
+    'global.anthropic.claude-haiku-5-5': {
+        inputPerMTok: 0.1,
+        outputPerMTok: 0.5,
+        thinkingPerMTok: 0.5,
+        cacheReadPerMTok: 0.01,
+        cacheCreationPerMTok: 0.125,
+        contextWindow: 1_000_000,
+        tierThreshold: 100_000,
+        tierInputPerMTok: 0.5,
+        tierOutputPerMTok: 2.5,
+        tierThinkingPerMTok: 2.5,
+        tierCacheReadPerMTok: 0.05,
+    },
+    // Fable 5.1 and Mythos 5.1: AWS model cards confirm these global IDs exist, but
+    // the pricing page lists no Global-tier row. Rates are DERIVED from the geo row
+    // via AWS's flat 1.1x geo premium; replace with published rates when AWS lists them.
+    'global.anthropic.claude-fable-5-1': {
+        inputPerMTok: 10,
+        outputPerMTok: 50,
+        cacheReadPerMTok: 0.25,
+        cacheCreationPerMTok: 12.5,
+        contextWindow: 1_000_000,
+    },
+    'global.anthropic.claude-mythos-5-1': {
+        inputPerMTok: 10,
+        outputPerMTok: 50,
+        cacheReadPerMTok: 0.25,
+        cacheCreationPerMTok: 12.5,
+        contextWindow: 1_000_000,
+    },
+    // Sonnet 4 is priced identically in the Geo and Global tiers on the AWS page.
+    'global.anthropic.claude-sonnet-4-20250514-v1:0': {
+        inputPerMTok: 3,
+        outputPerMTok: 15,
+        cacheReadPerMTok: 0.3,
+        cacheCreationPerMTok: 3.75,
+        contextWindow: 200_000,
+    },
     // Legacy Claude models via Bedrock cross-region inference
     // Cache rates verified against AWS Bedrock pricing page (2026-06-03) — but
     // against the WRONG table. AWS publishes two tiers for this model: the
     // standard $3/$15 tier (cache write $3.75, cache read $0.30 — used below,
-    // paired with the input/output rate this entry already had and this PR
-    // does not change) and a separate "Public Extended Access, Effective 1
+    // paired with this entry's standard input/output rate) and a separate "Public Extended Access, Effective 1
     // Dec 2025" tier at exactly 2x ($6/$30, cache write $7.50, cache read
     // $0.60) for accounts still on this model post-EOL. The 2026-06-03 check
     // paired the standard input/output with the extended-access cache rates
@@ -786,8 +1329,7 @@ export const DEFAULT_PRICING_TABLE = {
     // placeholders on 2026-08-14, not inferred), so it's internally
     // consistent again. If callers are actually on Public Extended Access —
     // undetectable from usage data alone — the whole entry (not just the
-    // cache rates) would need to move to the $6/$30 row instead; that's a
-    // separate decision this PR doesn't make.
+    // cache rates) would need to move to the $6/$30 row instead; not done here.
     'anthropic.claude-3-5-sonnet-20241022-v2:0': {
         inputPerMTok: 3,
         outputPerMTok: 15,
@@ -795,8 +1337,8 @@ export const DEFAULT_PRICING_TABLE = {
         cacheCreationPerMTok: 3.75,
         contextWindow: 200_000,
     },
-    // Cache pricing for this model IS now published on the Bedrock pricing
-    // page (it wasn't as of this file's last check) — same 1.25x-input /
+    // Cache pricing for this model is published on the Bedrock pricing
+    // page — same 1.25x-input /
     // 0.1x-input multiplier convention used everywhere else in this section.
     'anthropic.claude-3-5-haiku-20241022-v1:0': {
         inputPerMTok: 0.8,
@@ -825,7 +1367,65 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 0.6,
         contextWindow: 8_192,
     },
+    // Llama 3.1+ / 4 per-MTok rates are us-east-1 on-demand from AWS's pricing
+    // page; IDs and context windows confirmed against AWS model cards (2026-10-07).
+    'meta.llama4-maverick-17b-instruct-v1:0': {
+        inputPerMTok: 0.24,
+        outputPerMTok: 0.97,
+        contextWindow: 1_000_000,
+    },
+    'meta.llama4-scout-17b-instruct-v1:0': {
+        inputPerMTok: 0.17,
+        outputPerMTok: 0.66,
+        contextWindow: 10_000_000,
+    },
+    'meta.llama3-3-70b-instruct-v1:0': {
+        inputPerMTok: 0.72,
+        outputPerMTok: 0.72,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-2-1b-instruct-v1:0': {
+        inputPerMTok: 0.1,
+        outputPerMTok: 0.1,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-2-3b-instruct-v1:0': {
+        inputPerMTok: 0.15,
+        outputPerMTok: 0.15,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-2-11b-instruct-v1:0': {
+        inputPerMTok: 0.16,
+        outputPerMTok: 0.16,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-2-90b-instruct-v1:0': {
+        inputPerMTok: 0.72,
+        outputPerMTok: 0.72,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-1-8b-instruct-v1:0': {
+        inputPerMTok: 0.22,
+        outputPerMTok: 0.22,
+        contextWindow: 128_000,
+    },
+    'meta.llama3-1-70b-instruct-v1:0': {
+        inputPerMTok: 0.72,
+        outputPerMTok: 0.72,
+        contextWindow: 128_000,
+    },
     // Mistral via Bedrock
+    // Mistral 7B and Pixtral Large rates: us-east-1 on-demand, AWS page (2026-10-07).
+    'mistral.mistral-7b-instruct-v0:2': {
+        inputPerMTok: 0.15,
+        outputPerMTok: 0.2,
+        contextWindow: 32_000,
+    },
+    'mistral.pixtral-large-2502-v1:0': {
+        inputPerMTok: 2,
+        outputPerMTok: 6,
+        contextWindow: 128_000,
+    },
     'mistral.mistral-large-2402-v1:0': {
         inputPerMTok: 4,
         outputPerMTok: 12,
@@ -836,20 +1436,44 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 3,
         contextWindow: 32_000,
     },
-    // Amazon Nova
+    // Mistral Large 3 on Bedrock: $0.50/$1.50 per the AWS pricing page, 256K per its model card.
+    'mistral.mistral-large-3-675b-instruct': {
+        inputPerMTok: 0.5,
+        outputPerMTok: 1.5,
+        contextWindow: 256_000,
+    },
+    // Amazon Nova (Premier / 2 Lite added 2026-10-07). Nova Premier has no AWS model
+    // card but is still priced on the pricing page; its ID and 1M window are unconfirmed.
+    // Cache reads (25% of input) are from the Bedrock pricing page's Nova on-demand
+    // table (verified 2026-10-08); the prompt-caching docs list Micro/Lite/Pro/2 Lite
+    // as supported. Premier's cache support is unchecked. No cache-write charge seen.
+    'amazon.nova-premier-v1:0': {
+        inputPerMTok: 2.5,
+        outputPerMTok: 12.5,
+        contextWindow: 1_000_000,
+    },
+    'amazon.nova-2-lite-v1:0': {
+        inputPerMTok: 0.3,
+        outputPerMTok: 2.5,
+        cacheReadPerMTok: 0.075,
+        contextWindow: 1_000_000,
+    },
     'amazon.nova-pro-v1:0': {
         inputPerMTok: 0.8,
         outputPerMTok: 3.2,
+        cacheReadPerMTok: 0.2,
         contextWindow: 300_000,
     },
     'amazon.nova-lite-v1:0': {
         inputPerMTok: 0.06,
         outputPerMTok: 0.24,
+        cacheReadPerMTok: 0.015,
         contextWindow: 300_000,
     },
     'amazon.nova-micro-v1:0': {
         inputPerMTok: 0.035,
         outputPerMTok: 0.14,
+        cacheReadPerMTok: 0.00875,
         contextWindow: 128_000,
     },
     // ---- Mistral ----
@@ -873,6 +1497,24 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 1.5,
         contextWindow: 256_000,
     },
+    // Mistral Large 4 (docs.mistral.ai/models/mistral-large-4-0, v26.10, public
+    // preview since 2026-10-06; API names mistral-large-4 / mistral-large-4-0).
+    // Rates are the preview SALE price actually billed, verified 2026-10-07:
+    // $0.68 in / $2.09 out against a struck-through list of $1.36 / $4.18.
+    // TEMPORARY SALE PRICING, no published end date: re-verify at every pricing
+    // audit and revert to the list price ($1.36 / $4.18) once the sale ends,
+    // otherwise costs are under-reported by half. 'mistral-large-latest' above is
+    // deliberately unchanged: the docs don't say it now routes here.
+    'mistral-large-4': {
+        inputPerMTok: 0.68,
+        outputPerMTok: 2.09,
+        contextWindow: 1_000_000,
+    },
+    'mistral-large-4-0': {
+        inputPerMTok: 0.68,
+        outputPerMTok: 2.09,
+        contextWindow: 1_000_000,
+    },
     'mistral-small-latest': {
         inputPerMTok: 0.15,
         outputPerMTok: 0.6,
@@ -900,7 +1542,7 @@ export const DEFAULT_PRICING_TABLE = {
     },
     // Retired 2026-07-31 (deprecated 2026-05-22) — no longer billable via the
     // Mistral API. Retained for historical-cost backfill only, same
-    // convention as gemini-2.0-flash / command / command-light below.
+    // convention as gemini-2.0-flash (above) / command / command-light (below).
     // contextWindow unchanged (128k) — this predates the 256k bump the
     // -latest family got later; it was never repriced, only retired.
     'mistral-nemo': {
@@ -908,7 +1550,9 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 0.15,
         contextWindow: 131_072,
     },
-    // ---- Mistral (legacy — deprecated March 2025) ----
+    // ---- Mistral (legacy — retired 2025-03-30) ----
+    // Both are listed under "Deprecated & retired models" on docs.mistral.ai
+    // (checked 2026-10-07). Retained for historical-cost backfill.
     'open-mistral-7b': {
         inputPerMTok: 0.25,
         outputPerMTok: 0.25,
@@ -919,9 +1563,8 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 0.7,
         contextWindow: 32_000,
     },
-    // Note: codestral gets the SMALLER window (128k) — the generalist models
-    // above get the larger 256k window. This is backwards from what the table
-    // previously encoded (codestral had 256k, generalists had 131_072).
+    // Note: codestral's window (128k) is smaller than the 256k of the current
+    // -latest generalist models.
     'codestral-latest': {
         inputPerMTok: 0.3,
         outputPerMTok: 0.9,
@@ -1003,18 +1646,14 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 6,
         cacheReadPerMTok: 0.3,
         contextWindow: 500_000,
-        // ModelPricing has no tiered cache-read field, so the long-context
-        // tier's doubled cached-input rate ($0.6 vs the $0.3 base rate above)
-        // cannot be represented — the base (lower) cache rate is used at every
-        // tier. Not fixable without a schema change; known and accepted.
         tierThreshold: 200_000,
         tierMode: 'flat',
+        tierCacheReadPerMTok: 0.6,
         tierInputPerMTok: 4,
         tierOutputPerMTok: 12,
     },
-    // xAI's current flagship, GA (docs.x.ai marks it "Latest"/default model,
-    // no preview/beta disclaimer), sourced from docs.x.ai/docs/pricing and
-    // docs.x.ai/docs/models/grok-4.6 directly — same $2/$6 base and 200K flat
+    // Prior xAI flagship (superseded by grok-4.7), sourced from
+    // docs.x.ai/docs/pricing and docs.x.ai/docs/models/grok-4.6 — same $2/$6 base and 200K flat
     // long-context tier shape as grok-4.5, but a higher cached-input rate
     // ($0.5 base / $1 long-context tier, vs 4.5's $0.3/$0.6).
     'grok-4.6': {
@@ -1022,40 +1661,135 @@ export const DEFAULT_PRICING_TABLE = {
         outputPerMTok: 6,
         cacheReadPerMTok: 0.5,
         contextWindow: 500_000,
-        // Same known schema gap as grok-4.5: no tiered cache-read field, so the
-        // long-context tier's doubled cached-input rate ($1 vs the $0.5 base
-        // rate above) can't be represented — the base rate applies at every
-        // tier. Not fixable without a schema change; known and accepted.
         tierThreshold: 200_000,
         tierMode: 'flat',
+        tierCacheReadPerMTok: 1,
         tierInputPerMTok: 4,
         tierOutputPerMTok: 12,
     },
+    // Verified 2026-10-07 against docs.x.ai/developers/pricing. Every entry
+    // below has the same long-context (xAI: ">= 200k"; computeCost tiers
+    // strictly above 200_000 input tokens) flat tier as grok-4.5, with
+    // the cached-input rate doubled in that tier (tierCacheReadPerMTok).
+    // Grok 4.7 Fast and the 1.1x US-regional multiplier on the xAI page are
+    // separate SKUs (Cursor/Grok Build only; regional endpoint) — not modeled.
+    'grok-4.7': {
+        inputPerMTok: 2,
+        outputPerMTok: 6,
+        cacheReadPerMTok: 0.5,
+        contextWindow: 500_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 1,
+        tierInputPerMTok: 4,
+        tierOutputPerMTok: 12,
+    },
+    'grok-build-0.1': {
+        inputPerMTok: 1,
+        outputPerMTok: 2,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 256_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 2,
+        tierOutputPerMTok: 4,
+    },
+    'grok-4.3': {
+        inputPerMTok: 1.25,
+        outputPerMTok: 2.5,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 1_000_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 2.5,
+        tierOutputPerMTok: 5,
+    },
+    'grok-4.20-0309-reasoning': {
+        inputPerMTok: 1.25,
+        outputPerMTok: 2.5,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 1_000_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 2.5,
+        tierOutputPerMTok: 5,
+    },
+    'grok-4.20-0309-non-reasoning': {
+        inputPerMTok: 1.25,
+        outputPerMTok: 2.5,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 1_000_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 2.5,
+        tierOutputPerMTok: 5,
+    },
+    'grok-4.20-multi-agent-0309': {
+        inputPerMTok: 1.25,
+        outputPerMTok: 2.5,
+        cacheReadPerMTok: 0.2,
+        contextWindow: 1_000_000,
+        tierThreshold: 200_000,
+        tierMode: 'flat',
+        tierCacheReadPerMTok: 0.4,
+        tierInputPerMTok: 2.5,
+        tierOutputPerMTok: 5,
+    },
     // ---- Moonshot AI ----
-    // Sourced from GitHub Copilot's pricing page (see the file-level comment
-    // above) — Moonshot AI has no separate public per-token pricing page.
+    // Moonshot publishes direct pricing at platform.moonshot.ai/docs/pricing/chat
+    // (rates agree with GitHub Copilot's Moonshot table). K3 also bills cache
+    // writes separately by TTL (5min $3.00 / 1h $6.00, published on the
+    // context-caching guide, verified 2026-10-08). extractOpenAITokens() reads
+    // the usage object's single cache_write_tokens count into cacheCreationTokens
+    // and bills it at the 5min rate. The 1h TTL ($6.00) is not modeled: the TTL
+    // split is only in response headers, so 1h writes under-price by $3.00/MTok.
+    // GitHub's retirement table removes Kimi K2.7 Code from Copilot on 2026-10-02,
+    // but Moonshot's direct page still lists it, so the entry stays.
     'kimi-k2.7-code': {
         inputPerMTok: 0.95,
         outputPerMTok: 4,
         cacheReadPerMTok: 0.19,
-        contextWindow: 256_000,
+        contextWindow: 262_144,
     },
     'kimi-k3': {
         inputPerMTok: 3,
         outputPerMTok: 15,
         cacheReadPerMTok: 0.3,
-        contextWindow: 256_000,
+        cacheCreationPerMTok: 3,
+        contextWindow: 1_048_576,
+    },
+    // kimi-k2.7-code-highspeed and kimi-k2.6 verified 2026-10-07 against the
+    // same page (column headers: cache hit / cache miss / output).
+    'kimi-k2.7-code-highspeed': {
+        inputPerMTok: 1.9,
+        outputPerMTok: 8,
+        cacheReadPerMTok: 0.38,
+        contextWindow: 262_144,
+    },
+    'kimi-k2.6': {
+        inputPerMTok: 0.95,
+        outputPerMTok: 4,
+        cacheReadPerMTok: 0.16,
+        contextWindow: 262_144,
     },
     // ---- Microsoft (MAI-Code) ----
     // Sourced from GitHub Copilot's pricing page (see the file-level comment
     // above) — Microsoft has no separate public per-token pricing page for
     // the MAI-Code family.
+    // No longer on Copilot's pricing page (only mai-code-1.1-flash is, checked
+    // 2026-10-07) — likely superseded. Retained for historical-cost backfill.
     'mai-code-1-flash': {
         inputPerMTok: 0.75,
         outputPerMTok: 4.5,
         cacheReadPerMTok: 0.075,
         contextWindow: 128_000,
     },
+    // contextWindow unverified: neither Copilot nor Microsoft publishes one for the
+    // hosted model. Microsoft's launch post cites 256K for the on-device build only.
     'mai-code-1.1-flash': {
         inputPerMTok: 0.2,
         outputPerMTok: 1.2,
@@ -1065,7 +1799,9 @@ export const DEFAULT_PRICING_TABLE = {
     // ---- GitHub (Copilot fine-tuned) ----
     // Raptor mini is GitHub's own fine-tuned model, billed only through
     // Copilot — there is no vendor page to source it from other than Copilot's
-    // own pricing page (see the file-level comment above).
+    // own pricing page (see the file-level comment above). No longer listed
+    // there as of 2026-10-07 — likely delisted. Retained for historical-cost
+    // backfill.
     'raptor-mini': {
         inputPerMTok: 0.25,
         outputPerMTok: 2,