@molecule/api-resource-ai-models 1.0.1 → 1.0.2

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: 2026-08-04T01:49:08.449Z
+Generated: 2026-08-06T19:42:52.531Z
 -->
 
 # @molecule/api-resource-ai-models
@@ -563,6 +563,9 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
     opus-4-8 superseded by opus-5 at identical pricing but still served — it is
     the recommended refusal-fallback model; effort ladder on all three current
     models is low|medium|high|xhigh|max; budget_tokens 400s on 4.7+)
+    (re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
+    cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
+    models.dev first listed it 2026-08-06)
 - OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
   2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
   20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
@@ -591,9 +594,18 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
 - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo (unchanged;
   minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
 - Alibaba: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
-  (unchanged; qwen3.8-max-preview, 2026-07-19, is Token-Plan-subscriber-only
-  — not on the pay-as-you-go API, so it cannot be added yet; qwen3.7-max
-  currently runs a 50%-off promo — still billed here at list, $2.50/$7.50)
+  (qwen3.8-max GA'd 2026-08-03 on the pay-as-you-go international API and is
+  IN the catalog — verified 2026-08-04: flat $2/$6 per MTok on
+  help.aliyun.com/en/model-studio/model-pricing (Singapore International
+  CNY 14.988/44.965 at the same fixed conversion that maps qwen3.7-max's
+  CNY 18.736/56.207 to its $2.50/$7.50 list), 1M ctx, hybrid thinking,
+  tools. Cache rates come from the ZH context-cache doc
+  (help.aliyun.com/zh/model-studio/context-cache), which lists qwen3.8-max
+  as supported in every region under the unconditional standard table
+  (implicit: hit 20% of input, creation 100%; explicit: hit 10%, creation
+  125%) — the EN edition of that doc simply lags (zero qwen3.8 mentions),
+  which an earlier pass misread as "excepted/console-only". qwen3.7-max
+  still runs its 50%-off promo — billed here at list, $2.50/$7.50)
 - Zhipu: https://docs.z.ai/guides/overview/pricing (unchanged; glm-5.2 is
   the newest — "GLM-5.3/5.5" rumors have no released ids as of 2026-07-28)
 

package/dist/models.d.ts

@@ -36,6 +36,9 @@ import type { ModelDefinition } from './types.js';
  *   opus-4-8 superseded by opus-5 at identical pricing but still served — it is
  *   the recommended refusal-fallback model; effort ladder on all three current
  *   models is low|medium|high|xhigh|max; budget_tokens 400s on 4.7+)
+ *   (re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
+ *   cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
+ *   models.dev first listed it 2026-08-06)
  * - OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
  *   2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
  *   20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
@@ -64,9 +67,18 @@ import type { ModelDefinition } from './types.js';
  * - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo (unchanged;
  *   minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
  * - Alibaba: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
- *   (unchanged; qwen3.8-max-preview, 2026-07-19, is Token-Plan-subscriber-only
- *   — not on the pay-as-you-go API, so it cannot be added yet; qwen3.7-max
- *   currently runs a 50%-off promo — still billed here at list, $2.50/$7.50)
+ *   (qwen3.8-max GA'd 2026-08-03 on the pay-as-you-go international API and is
+ *   IN the catalog — verified 2026-08-04: flat $2/$6 per MTok on
+ *   help.aliyun.com/en/model-studio/model-pricing (Singapore International
+ *   CNY 14.988/44.965 at the same fixed conversion that maps qwen3.7-max's
+ *   CNY 18.736/56.207 to its $2.50/$7.50 list), 1M ctx, hybrid thinking,
+ *   tools. Cache rates come from the ZH context-cache doc
+ *   (help.aliyun.com/zh/model-studio/context-cache), which lists qwen3.8-max
+ *   as supported in every region under the unconditional standard table
+ *   (implicit: hit 20% of input, creation 100%; explicit: hit 10%, creation
+ *   125%) — the EN edition of that doc simply lags (zero qwen3.8 mentions),
+ *   which an earlier pass misread as "excepted/console-only". qwen3.7-max
+ *   still runs its 50%-off promo — billed here at list, $2.50/$7.50)
  * - Zhipu: https://docs.z.ai/guides/overview/pricing (unchanged; glm-5.2 is
  *   the newest — "GLM-5.3/5.5" rumors have no released ids as of 2026-07-28)
  *

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;AAEjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiEG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EA6oCnC,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;AAEjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA6EG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EAmtCnC,CAAA"}

package/dist/models.js

@@ -35,6 +35,9 @@
  *   opus-4-8 superseded by opus-5 at identical pricing but still served — it is
  *   the recommended refusal-fallback model; effort ladder on all three current
  *   models is low|medium|high|xhigh|max; budget_tokens 400s on 4.7+)
+ *   (re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
+ *   cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
+ *   models.dev first listed it 2026-08-06)
  * - OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
  *   2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
  *   20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
@@ -63,9 +66,18 @@
  * - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo (unchanged;
  *   minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
  * - Alibaba: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
- *   (unchanged; qwen3.8-max-preview, 2026-07-19, is Token-Plan-subscriber-only
- *   — not on the pay-as-you-go API, so it cannot be added yet; qwen3.7-max
- *   currently runs a 50%-off promo — still billed here at list, $2.50/$7.50)
+ *   (qwen3.8-max GA'd 2026-08-03 on the pay-as-you-go international API and is
+ *   IN the catalog — verified 2026-08-04: flat $2/$6 per MTok on
+ *   help.aliyun.com/en/model-studio/model-pricing (Singapore International
+ *   CNY 14.988/44.965 at the same fixed conversion that maps qwen3.7-max's
+ *   CNY 18.736/56.207 to its $2.50/$7.50 list), 1M ctx, hybrid thinking,
+ *   tools. Cache rates come from the ZH context-cache doc
+ *   (help.aliyun.com/zh/model-studio/context-cache), which lists qwen3.8-max
+ *   as supported in every region under the unconditional standard table
+ *   (implicit: hit 20% of input, creation 100%; explicit: hit 10%, creation
+ *   125%) — the EN edition of that doc simply lags (zero qwen3.8 mentions),
+ *   which an earlier pass misread as "excepted/console-only". qwen3.7-max
+ *   still runs its 50%-off promo — billed here at list, $2.50/$7.50)
  * - Zhipu: https://docs.z.ai/guides/overview/pricing (unchanged; glm-5.2 is
  *   the newest — "GLM-5.3/5.5" rumors have no released ids as of 2026-07-28)
  *
@@ -213,6 +225,40 @@ export const MODELS = [
         cacheWritePricePerMTok: 3.75,
         knowledgeCutoff: '2026-01-01',
     },
+    {
+        id: 'claude-opus-4-7',
+        provider: 'anthropic',
+        label: 'Claude Opus 4.7',
+        description: 'Older Opus — long-horizon agentic work, knowledge work & vision',
+        contextWindow: 1_000_000,
+        maxOutputTokens: 128_000,
+        supportsThinking: true,
+        thinkingBudgetTokens: 16_000,
+        thinkingConfigurable: true,
+        supportedEffortLevels: ['low', 'medium', 'high', 'xhigh', 'max'],
+        defaultEffortLevel: 'high',
+        // Adaptive thinking only — budget_tokens is REJECTED (400); xhigh effort
+        // debuted on this model. Unlike opus-5, omitting `thinking` runs WITHOUT
+        // thinking — set {type:"adaptive"} explicitly. First model on the new
+        // tokenizer (~30% more tokens than 4.6 for the same text).
+        supportsVision: true,
+        supportsPromptCaching: true,
+        supportsTools: true,
+        webSearchToolType: 'web_search_20260209',
+        codeExecutionToolType: 'code_execution_20250825',
+        webFetchToolType: 'web_fetch_20260209',
+        inputPricePerMTok: 5,
+        outputPricePerMTok: 25,
+        // Anthropic 5-minute prompt cache: read 0.1× input, write 1.25× input.
+        cacheReadPricePerMTok: 0.5,
+        cacheWritePricePerMTok: 6.25,
+        knowledgeCutoff: '2026-01-01',
+        // Superseded by claude-opus-4-8 (launched 2026-05-28) at identical pricing;
+        // still Active upstream (deprecations page 2026-08-06: retires no sooner
+        // than 2027-04-16). Selectable under "Older models". NO fast mode —
+        // speed:"fast" on 4.7 returns an error (pricing page, fast-mode section).
+        deprecatedAt: '2026-05-28',
+    },
     {
         id: 'claude-opus-4-6',
         provider: 'anthropic',
@@ -1097,6 +1143,42 @@ export const MODELS = [
     // Prices are DashScope international list rates (the bond calls DashScope,
     // not OpenRouter; a 50%-off promo currently applies — billed at list).
     // ---------------------------------------------------------------------------
+    // qwen3.8-max (GA 2026-08-03) succeeds qwen3.7-max as the agentic flagship,
+    // priced BELOW it at $2/$6 (intl CNY 14.988/44.965, same fixed conversion).
+    // Same hybrid thinking mechanism as the 3.7 series (enable_thinking default
+    // ON + thinking_budget; preserve_thinking supported). Context cache uses the
+    // standard implicit rates — see the Sources block for the ZH-doc citation.
+    {
+        id: 'qwen3.8-max',
+        provider: 'alibaba',
+        label: 'Qwen3.8 Max',
+        description: 'Alibaba agentic flagship — 1M context, hybrid thinking',
+        contextWindow: 1_000_000,
+        // Alibaba's public pages don't state a max-output figure; models.dev says
+        // 131,072 — kept at the 3.7-max figure until the provider publishes one
+        // (understating only shortens completions; overstating would error).
+        maxOutputTokens: 65_536,
+        supportsThinking: true,
+        thinkingBudgetTokens: 8_000,
+        thinkingConfigurable: true,
+        supportedEffortLevels: ['4K', '8K', '16K', '32K'],
+        defaultEffortLevel: '8K',
+        effortBudgetTokens: { '4K': 4000, '8K': 8000, '16K': 16000, '32K': 32000 },
+        // models.dev claims image+video input, but Alibaba's own model catalog
+        // lists qwen3.8-max under text generation (VL remains a separate line) —
+        // false until the provider's page says otherwise.
+        supportsVision: false,
+        supportsPromptCaching: true,
+        supportsTools: true,
+        inputPricePerMTok: 2,
+        outputPricePerMTok: 6,
+        // Implicit context cache: read = 20% of input, no write premium.
+        cacheReadPricePerMTok: 0.4,
+        cacheWritePricePerMTok: 2,
+        regions: ['us', 'cn'],
+        // Not published by Alibaba — best-effort estimate.
+        knowledgeCutoff: '2026-04-01',
+    },
     {
         id: 'qwen3.7-max',
         provider: 'alibaba',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@molecule/api-resource-ai-models",
-  "version": "1.0.1",
+  "version": "1.0.2",
   "description": "AI model catalog — server-side source of truth plus an authentication-gated discovery endpoint",
   "type": "module",
   "main": "dist/index.js",