@molecule/api-resource-ai-models 1.4.0 → 1.5.1

package/README.md

@@ -3,7 +3,7 @@ AUTO-GENERATED — DO NOT EDIT THIS FILE.
 Generated by `mlcl sync-docs` from the package's src/index.ts JSDoc + mlcl/registry.json.
 Edits here are overwritten on the next commit (molecule's pre-commit hook regenerates).
 To change this document, edit the module-level JSDoc in src/index.ts.
-Generated: 2026-09-01T19:40:01.931Z
+Generated: 2026-09-06T11:38:29.688Z
 -->
 
 # @molecule/api-resource-ai-models
@@ -846,13 +846,22 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
   at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
   never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
   2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
+  (re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
+  3.7-flash, which the models page now calls "previous-generation". THIRD id
+  on the same flash-tier promo row, verbatim "$0.75 through December 31,
+  2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
+  $0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
+  Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
+  low|medium|high, vision, tools, caching, search grounding, code execution,
+  url context — identical surface to 3.7-flash. Knowledge cutoff is published
+  by neither Google nor models.dev for this id)
 - xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
   (grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
   not modeled; reasoning_effort low|medium|high default high, image input;
   grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
   grok-code-fast-1 no longer listed — retires 2026-08-15)
 - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
-  2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
+  2026-09-06; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
   never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
   has LANDED and is folded into the base fields, along with the peak-hour 2×
   the same card introduced; every rate re-read on the card 2026-08-31 and
@@ -861,7 +870,10 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
   where on 2026-08-18 it carried no day qualifier at all, so the windows are
   `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
   also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
-  deliberately not catalogued.)
+  deliberately not catalogued. The 2026-09-06 re-read found every native rate
+  unchanged; what moved was the US re-host — DeepInfra cut
+  `DeepSeek-V4-Flash-0731` from $0.08 to $0.06 input, cache read $0.016 to
+  $0.015, output held at $0.18. See that entry's `regionPricing`.)
 - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
   the US re-host (kimi-k3 flagship 2026-07-16
   — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with

package/dist/models.d.ts

@@ -114,13 +114,22 @@ import type { ModelDefinition } from './types.js';
  *   at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
  *   never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
  *   2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
+ *   (re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
+ *   3.7-flash, which the models page now calls "previous-generation". THIRD id
+ *   on the same flash-tier promo row, verbatim "$0.75 through December 31,
+ *   2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
+ *   $0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
+ *   Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
+ *   low|medium|high, vision, tools, caching, search grounding, code execution,
+ *   url context — identical surface to 3.7-flash. Knowledge cutoff is published
+ *   by neither Google nor models.dev for this id)
  * - xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
  *   (grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
  *   not modeled; reasoning_effort low|medium|high default high, image input;
  *   grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
  *   grok-code-fast-1 no longer listed — retires 2026-08-15)
  * - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
- *   2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
+ *   2026-09-06; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
  *   never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
  *   has LANDED and is folded into the base fields, along with the peak-hour 2×
  *   the same card introduced; every rate re-read on the card 2026-08-31 and
@@ -129,7 +138,10 @@ import type { ModelDefinition } from './types.js';
  *   where on 2026-08-18 it carried no day qualifier at all, so the windows are
  *   `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
  *   also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
- *   deliberately not catalogued.)
+ *   deliberately not catalogued. The 2026-09-06 re-read found every native rate
+ *   unchanged; what moved was the US re-host — DeepInfra cut
+ *   `DeepSeek-V4-Flash-0731` from $0.08 to $0.06 input, cache read $0.016 to
+ *   $0.015, output held at $0.18. See that entry's `regionPricing`.)
  * - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
  *   the US re-host (kimi-k3 flagship 2026-07-16
  *   — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with

package/dist/models.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAA;AAQjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAwLG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EAupDnC,CAAA"}
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAA;AAQjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoMG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EA6sDnC,CAAA"}

package/dist/models.js

@@ -118,13 +118,22 @@ const WEEKDAYS_UTC = [1, 2, 3, 4, 5];
  *   at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
  *   never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
  *   2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
+ *   (re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
+ *   3.7-flash, which the models page now calls "previous-generation". THIRD id
+ *   on the same flash-tier promo row, verbatim "$0.75 through December 31,
+ *   2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
+ *   $0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
+ *   Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
+ *   low|medium|high, vision, tools, caching, search grounding, code execution,
+ *   url context — identical surface to 3.7-flash. Knowledge cutoff is published
+ *   by neither Google nor models.dev for this id)
  * - xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
  *   (grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
  *   not modeled; reasoning_effort low|medium|high default high, image input;
  *   grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
  *   grok-code-fast-1 no longer listed — retires 2026-08-15)
  * - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
- *   2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
+ *   2026-09-06; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
  *   never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
  *   has LANDED and is folded into the base fields, along with the peak-hour 2×
  *   the same card introduced; every rate re-read on the card 2026-08-31 and
@@ -133,7 +142,10 @@ const WEEKDAYS_UTC = [1, 2, 3, 4, 5];
  *   where on 2026-08-18 it carried no day qualifier at all, so the windows are
  *   `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
  *   also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
- *   deliberately not catalogued.)
+ *   deliberately not catalogued. The 2026-09-06 re-read found every native rate
+ *   unchanged; what moved was the US re-host — DeepInfra cut
+ *   `DeepSeek-V4-Flash-0731` from $0.08 to $0.06 input, cache read $0.016 to
+ *   $0.015, output held at $0.18. See that entry's `regionPricing`.)
  * - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
  *   the US re-host (kimi-k3 flagship 2026-07-16
  *   — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
@@ -776,11 +788,48 @@ export const MODELS = [
     // replace outright: the google bond has never been implemented/wired, so no
     // historical usage can reference the old ids.
     // ---------------------------------------------------------------------------
+    {
+        id: 'gemini-3.8-flash',
+        provider: 'google',
+        label: 'Gemini 3.8 Flash',
+        description: 'Google agentic flagship — long-horizon engineering & autonomous agents',
+        // Verified against /docs/models/gemini-3.8-flash (2026-09-02) — same limits
+        // and capability surface as 3.7-flash.
+        contextWindow: 1_048_576,
+        maxOutputTokens: 65_536,
+        supportsThinking: true,
+        thinkingBudgetTokens: 10_000,
+        thinkingConfigurable: true,
+        // thinking_level low|medium|high — minimal NOT supported on this model.
+        supportedEffortLevels: ['low', 'medium', 'high'],
+        defaultEffortLevel: 'medium',
+        supportsVision: true,
+        supportsPromptCaching: true,
+        supportsTools: true,
+        webSearchToolType: 'google_search',
+        codeExecutionToolType: 'code_execution',
+        webFetchToolType: 'url_context',
+        // "New Stable" 2026-09-02. LIST price $1.50/$7.50 — same card as 3.7- and
+        // 3.6-flash, and on the same flash-tier launch promo ($0.75/$3.75, cache
+        // read $0.075) through 2026-12-31; billed here at standard list so metering
+        // never under-charges (identical policy and expiry to the two entries
+        // below — see the matching KNOWN_DIVERGENCES entry in
+        // scripts/check-model-freshness.mjs).
+        inputPricePerMTok: 1.5,
+        outputPricePerMTok: 7.5,
+        // Gemini context cache: read $0.15/M (0.1× input), no write premium
+        // (storage billed separately per hour — not modeled).
+        cacheReadPricePerMTok: 0.15,
+        cacheWritePricePerMTok: 1.5,
+        // Published by neither Google's model page nor models.dev for this id —
+        // carried from 3.7-flash (same generation) as a best-effort estimate.
+        knowledgeCutoff: '2026-03-01',
+    },
     {
         id: 'gemini-3.7-flash',
         provider: 'google',
         label: 'Gemini 3.7 Flash',
-        description: 'Google agentic flagship — complex coding & multi-step execution',
+        description: 'Previous Google agentic flagship — complex coding & multi-step execution',
         // Verified against /docs/models/gemini-3.7-flash (2026-08-13).
         contextWindow: 1_048_576,
         maxOutputTokens: 65_536,
@@ -810,6 +859,11 @@ export const MODELS = [
         cacheWritePricePerMTok: 1.5,
         // Not on Google's docs — models.dev reports 2026-03 (lead, not authority).
         knowledgeCutoff: '2026-03-01',
+        // Superseded by gemini-3.8-flash (2026-09-02) — same flash tier, same list
+        // price; Google's own models page now calls 3.7 "previous-generation".
+        // Still served upstream, so it stays priceable.
+        deprecatedAt: '2026-09-02',
+        supersededBy: 'gemini-3.8-flash',
     },
     {
         id: 'gemini-3.6-flash',
@@ -850,9 +904,11 @@ export const MODELS = [
         knowledgeCutoff: '2026-01-01',
         // Superseded by gemini-3.7-flash (2026-08-13) — same flash tier, same list
         // price; Google's own models page now calls 3.6 "previous-generation".
-        // Still served upstream, so it stays priceable.
+        // Still served upstream, so it stays priceable. Re-pointed to
+        // gemini-3.8-flash on 2026-09-02 when 3.7 was itself deprecated —
+        // supersededBy must target a SELECTABLE model, never a chain.
         deprecatedAt: '2026-08-13',
-        supersededBy: 'gemini-3.7-flash',
+        supersededBy: 'gemini-3.8-flash',
     },
     {
         id: 'gemini-3.5-flash',
@@ -883,11 +939,11 @@ export const MODELS = [
         cacheWritePricePerMTok: 1.5,
         knowledgeCutoff: '2025-01-01',
         // Superseded within the flash tier (first by 3.6-flash on 2026-07-21, now
-        // pointed one hop to gemini-3.7-flash — supersededBy must target a
+        // pointed one hop to gemini-3.8-flash — supersededBy must target a
         // SELECTABLE model, never a chain). Still served upstream, so it stays
         // priceable.
         deprecatedAt: '2026-07-21',
-        supersededBy: 'gemini-3.7-flash',
+        supersededBy: 'gemini-3.8-flash',
     },
     {
         id: 'gemini-3.1-pro-preview',
@@ -1221,23 +1277,29 @@ export const MODELS = [
         // US (DeepInfra) DEFAULT as of 2026-08-16 — flipped from CN when DeepSeek's
         // rise landed (owner decision 2026-08-14). CN was cheaper on real traffic
         // only because of its cache reads; the rise takes those from $0.0028 to
-        // $0.007 (peak $0.014) against DeepInfra's flat $0.016, which is no longer
-        // enough to carry the 1.6-3.1x it now loses on fresh input and output. On
+        // $0.007 (peak $0.014) against DeepInfra's flat $0.015, which is no longer
+        // enough to carry the 3.7x CN now loses on BOTH fresh input (0.22 vs 0.06)
+        // and output (0.66 vs 0.18). On
         // the agentic mix this model actually serves (~94% cache hits) US is
-        // cheaper at EVERY hour: 0.114c/turn flat vs 0.152c off-peak and 0.303c at
+        // cheaper at EVERY hour: 0.103c/turn flat vs 0.152c off-peak and 0.303c at
         // peak. It is also flat-rate, so free-tier cost stops varying by Beijing
         // business hours. Re-derive if the cache-hit ratio drops much below ~90%,
         // where CN's cheaper reads start winning again. This deliberately splits
         // the plan/execute pair across regions — Pro stays CN because its US
         // re-host is ~2.3x its own native rate even after the rise.
         regions: ['us', 'cn'],
-        // US = DeepInfra, verified 2026-08-13 against the id the bond actually
+        // US = DeepInfra, verified 2026-09-06 against the id the bond actually
         // sends: `deepseek-ai/DeepSeek-V4-Flash-0731`, the official release that
         // supersedes the preview weights still served under the un-dated id
-        // (cents_per_input_token 0.000008, cents_per_output_token 0.000018,
-        // rate_per_input_token_cached 0.2 → cache read = 0.2 × input).
+        // (cents_per_input_token 0.000006, cents_per_output_token 0.000018,
+        // rate_per_input_token_cached 0.25 → cache read = 0.25 × input).
+        // DeepInfra CUT the 0731 input rate 0.08 → 0.06 (cache read 0.016 →
+        // 0.015); output held at 0.18. The un-dated `DeepSeek-V4-Flash` id moved
+        // the other way (0.09 input, cache 0.2× = 0.018), so the two ids no longer
+        // price alike — this entry tracks the dated one the modelMap sends, and the
+        // gap widens the case for the US default this model already carries.
         regionPricing: {
-            us: { inputPricePerMTok: 0.08, outputPricePerMTok: 0.18, cacheReadPricePerMTok: 0.016 },
+            us: { inputPricePerMTok: 0.06, outputPricePerMTok: 0.18, cacheReadPricePerMTok: 0.015 },
         },
         // Peak-hour surcharge live since 2026-08-16T16:00Z, Mon-Fri (see
         // deepseek-v4-pro). It applies to the NATIVE CN card only — this model
@@ -1743,9 +1805,13 @@ export const MODELS = [
         // GLM context cache: read ≈0.19× input, no write premium.
         cacheReadPricePerMTok: 0.26,
         cacheWritePricePerMTok: 1.4,
-        // No US re-host exists — DeepInfra serves GLM-5.2 and GLM-5.3-Flash but
-        // returns "model not found" for zai-org/GLM-5.3 (checked 2026-08-26), so
-        // this is pinned to the native host and bills the list card above.
+        // Pinned to the native host, billing the list card above. DeepInfra has
+        // since STARTED serving zai-org/GLM-5.3 (it 404'd when this was cataloged
+        // on 2026-08-26; live 2026-09-02 at $1.20/$4.00, cache read 0.1× = $0.12,
+        // i.e. below native). Adding 'us' here is not a catalog-only edit — it
+        // needs a matching entry in molecule-dev's region-model-maps.ts and a
+        // regionPricing block, moved together, or dispatch sends the canonical id
+        // and 404s. Left as a human decision.
         regions: ['cn'],
         // Same base weights as glm-5.2, so the same best-effort estimate — Z.ai
         // publishes no cutoff.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@molecule/api-resource-ai-models",
-  "version": "1.4.0",
+  "version": "1.5.1",
   "description": "AI model catalog — server-side source of truth plus an authentication-gated discovery endpoint",
   "type": "module",
   "main": "dist/index.js",