@molecule/api-resource-ai-models 1.16.1 → 1.17.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/models.d.ts +26 -2
- package/dist/models.d.ts.map +1 -1
- package/dist/models.js +79 -7
- package/package.json +1 -1
package/dist/models.d.ts
CHANGED
|
@@ -273,8 +273,19 @@ import type { ModelDefinition } from './types.js';
|
|
|
273
273
|
* labeled Kimi-K3 — so kimi-k2.7-code is native-only (regions ['cn']) and
|
|
274
274
|
* its US modelMap entry is gone. kimi-k3's US re-host is healthy, rates
|
|
275
275
|
* unchanged.)
|
|
276
|
-
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
277
|
-
* minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
276
|
+
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
277
|
+
* (unchanged; minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
278
|
+
* (re-verified 2026-10-19 across platform.minimax.io — pricing-paygo,
|
|
279
|
+
* models-intro, the text-generation guide, the text-chat-openai reference,
|
|
280
|
+
* faq/models, the model release notes and the prompt-caching page: M3 rates
|
|
281
|
+
* unchanged at $0.30/$1.20/cache-read $0.06; MiniMax-M3.1-Flash (released
|
|
282
|
+
* 2026-10-19) added at $0.15/$0.60/cache-read $0.015, 1M ctx, multimodal,
|
|
283
|
+
* ALWAYS thinks (thinking.type "disabled" → 400) with depth via
|
|
284
|
+
* reasoning_effort low|medium|high|xhigh|max (API default max); no
|
|
285
|
+
* cache-write line item exists for ANY MiniMax model (billing components
|
|
286
|
+
* are new input + cache hit + output only). M2.5/M2.7 closed to new users
|
|
287
|
+
* 2026-10-19 and retire 2026-11-18, with M2.x plan traffic auto-routed to
|
|
288
|
+
* M3.1-Flash.)
|
|
278
289
|
* - Alibaba: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
|
|
279
290
|
* (qwen3.8-max GA'd 2026-08-03 on the pay-as-you-go international API and is
|
|
280
291
|
* IN the catalog — verified 2026-08-04: flat $2/$6 per MTok on
|
|
@@ -352,6 +363,19 @@ import type { ModelDefinition } from './types.js';
|
|
|
352
363
|
* output $0.50, cache read 0.2× input); deployment not deprecated.
|
|
353
364
|
* 'us' stays selectable at that same card, like deepseek-flash's cn-first
|
|
354
365
|
* default. glm-5.3 / glm-5.3-flashx unchanged, cn-only.)
|
|
366
|
+
* (re-verified 2026-10-20: every GLM rate on the pricing page is UNCHANGED —
|
|
367
|
+
* glm-5.3 / glm-5.2 $1.40/$4.40 (cached $0.26), glm-5.3-flashx $0.37/$1.25
|
|
368
|
+
* (cached $0.075), glm-5.3-flash $0.15/$0.50 (cached $0.03), glm-5 $1/$3.20
|
|
369
|
+
* (cached $0.20); the page states "All prices are in USD", cached-input
|
|
370
|
+
* storage "Limited-time Free", no other promo footnote. The 2026-10-20 06:35
|
|
371
|
+
* alert's 15 "price drifts" were FALSE: models.dev added/changed its
|
|
372
|
+
* `zhipuai` entry — open.bigmodel.cn's CNY card converted to USD (glm-5.3
|
|
373
|
+
* $1.19516/$4.18306 = ¥8/¥28 input/output, ¥2 cached, converted at ≈6.694;
|
|
374
|
+
* same relation holds for every other GLM entry) — and check-model-freshness
|
|
375
|
+
* compared this card against it. The zhipu bond bills z.ai
|
|
376
|
+
* (https://api.z.ai/api/paas), whose models.dev mirror is `zai` and carries
|
|
377
|
+
* this card exactly; PROVIDER_MAP in check-model-freshness.mjs now points
|
|
378
|
+
* there. No catalog price changed.)
|
|
355
379
|
*
|
|
356
380
|
* Knowledge-cutoff dates on non-Anthropic entries are best-effort estimates
|
|
357
381
|
* where the provider doesn't publish one; the provider sources above verify
|
package/dist/models.d.ts.map
CHANGED
|
@@ -1 +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;AAwDjD
|
|
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;AAwDjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoXG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EAi8EnC,CAAA"}
|
package/dist/models.js
CHANGED
|
@@ -322,8 +322,19 @@ const CHINA_PUBLIC_HOLIDAYS = [
|
|
|
322
322
|
* labeled Kimi-K3 — so kimi-k2.7-code is native-only (regions ['cn']) and
|
|
323
323
|
* its US modelMap entry is gone. kimi-k3's US re-host is healthy, rates
|
|
324
324
|
* unchanged.)
|
|
325
|
-
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
326
|
-
* minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
325
|
+
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
326
|
+
* (unchanged; minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
327
|
+
* (re-verified 2026-10-19 across platform.minimax.io — pricing-paygo,
|
|
328
|
+
* models-intro, the text-generation guide, the text-chat-openai reference,
|
|
329
|
+
* faq/models, the model release notes and the prompt-caching page: M3 rates
|
|
330
|
+
* unchanged at $0.30/$1.20/cache-read $0.06; MiniMax-M3.1-Flash (released
|
|
331
|
+
* 2026-10-19) added at $0.15/$0.60/cache-read $0.015, 1M ctx, multimodal,
|
|
332
|
+
* ALWAYS thinks (thinking.type "disabled" → 400) with depth via
|
|
333
|
+
* reasoning_effort low|medium|high|xhigh|max (API default max); no
|
|
334
|
+
* cache-write line item exists for ANY MiniMax model (billing components
|
|
335
|
+
* are new input + cache hit + output only). M2.5/M2.7 closed to new users
|
|
336
|
+
* 2026-10-19 and retire 2026-11-18, with M2.x plan traffic auto-routed to
|
|
337
|
+
* M3.1-Flash.)
|
|
327
338
|
* - Alibaba: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
|
|
328
339
|
* (qwen3.8-max GA'd 2026-08-03 on the pay-as-you-go international API and is
|
|
329
340
|
* IN the catalog — verified 2026-08-04: flat $2/$6 per MTok on
|
|
@@ -401,6 +412,19 @@ const CHINA_PUBLIC_HOLIDAYS = [
|
|
|
401
412
|
* output $0.50, cache read 0.2× input); deployment not deprecated.
|
|
402
413
|
* 'us' stays selectable at that same card, like deepseek-flash's cn-first
|
|
403
414
|
* default. glm-5.3 / glm-5.3-flashx unchanged, cn-only.)
|
|
415
|
+
* (re-verified 2026-10-20: every GLM rate on the pricing page is UNCHANGED —
|
|
416
|
+
* glm-5.3 / glm-5.2 $1.40/$4.40 (cached $0.26), glm-5.3-flashx $0.37/$1.25
|
|
417
|
+
* (cached $0.075), glm-5.3-flash $0.15/$0.50 (cached $0.03), glm-5 $1/$3.20
|
|
418
|
+
* (cached $0.20); the page states "All prices are in USD", cached-input
|
|
419
|
+
* storage "Limited-time Free", no other promo footnote. The 2026-10-20 06:35
|
|
420
|
+
* alert's 15 "price drifts" were FALSE: models.dev added/changed its
|
|
421
|
+
* `zhipuai` entry — open.bigmodel.cn's CNY card converted to USD (glm-5.3
|
|
422
|
+
* $1.19516/$4.18306 = ¥8/¥28 input/output, ¥2 cached, converted at ≈6.694;
|
|
423
|
+
* same relation holds for every other GLM entry) — and check-model-freshness
|
|
424
|
+
* compared this card against it. The zhipu bond bills z.ai
|
|
425
|
+
* (https://api.z.ai/api/paas), whose models.dev mirror is `zai` and carries
|
|
426
|
+
* this card exactly; PROVIDER_MAP in check-model-freshness.mjs now points
|
|
427
|
+
* there. No catalog price changed.)
|
|
404
428
|
*
|
|
405
429
|
* Knowledge-cutoff dates on non-Anthropic entries are best-effort estimates
|
|
406
430
|
* where the provider doesn't publish one; the provider sources above verify
|
|
@@ -2322,12 +2346,55 @@ export const MODELS = [
|
|
|
2322
2346
|
// MiniMax
|
|
2323
2347
|
// Verified: https://platform.minimax.io/docs/guides/models-intro
|
|
2324
2348
|
// https://platform.minimax.io/docs/guides/pricing-paygo.md
|
|
2349
|
+
// https://platform.minimax.io/docs/api-reference/text-chat-openai.md
|
|
2325
2350
|
// minimax-m3 (2026-06-01) is the flagship: 1M context (≥512K guaranteed),
|
|
2326
2351
|
// multimodal (text+image+video in), thinking adaptive|disabled (no budget /
|
|
2327
2352
|
// effort levels). Run non-thinking like the DeepSeek executors — with
|
|
2328
2353
|
// thinking on, reasoning must be carried through tool loops. m2.7 thinking
|
|
2329
|
-
// CANNOT be disabled upstream.
|
|
2354
|
+
// CANNOT be disabled upstream. m3.1-flash (2026-10-19) also cannot disable
|
|
2355
|
+
// thinking, but TUNES depth via reasoning_effort (low…max) — it is the
|
|
2356
|
+
// successor of the M2.x line (MiniMax routes retired M2.x plan traffic to
|
|
2357
|
+
// it), not of M3, which MiniMax still lists as current.
|
|
2330
2358
|
// ---------------------------------------------------------------------------
|
|
2359
|
+
{
|
|
2360
|
+
id: 'minimax-m3.1-flash',
|
|
2361
|
+
provider: 'minimax',
|
|
2362
|
+
label: 'MiniMax M3.1 Flash',
|
|
2363
|
+
description: 'Fast multimodal MiniMax — always-on thinking, tunable effort',
|
|
2364
|
+
contextWindow: 1_000_000,
|
|
2365
|
+
// Upstream recommends 128K max_completion_tokens (hard max 512K), same as
|
|
2366
|
+
// M3 (text-chat-openai reference).
|
|
2367
|
+
maxOutputTokens: 131_072,
|
|
2368
|
+
// ALWAYS thinks: thinking {type:"disabled"} → 400 "requires adaptive
|
|
2369
|
+
// thinking", and reasoning_effort "none" is rejected too (docs + live
|
|
2370
|
+
// probe 2026-10-19). Depth is tuned with reasoning_effort
|
|
2371
|
+
// low|medium|high|xhigh|max — the bond sends adaptive + the resolved
|
|
2372
|
+
// level and never a disabled flag.
|
|
2373
|
+
supportsThinking: true,
|
|
2374
|
+
thinkingBudgetTokens: 16_000,
|
|
2375
|
+
thinkingConfigurable: true,
|
|
2376
|
+
supportedEffortLevels: ['low', 'medium', 'high', 'xhigh', 'max'],
|
|
2377
|
+
// API default is max; MiniMax's own latency guidance is to LOWER the
|
|
2378
|
+
// effort level rather than wish thinking off — high is the balanced
|
|
2379
|
+
// default, xhigh/max for capability-sensitive turns.
|
|
2380
|
+
defaultEffortLevel: 'high',
|
|
2381
|
+
supportsVision: true,
|
|
2382
|
+
supportsPromptCaching: true,
|
|
2383
|
+
supportsTools: true,
|
|
2384
|
+
inputPricePerMTok: 0.15,
|
|
2385
|
+
outputPricePerMTok: 0.6,
|
|
2386
|
+
cacheReadPricePerMTok: 0.015,
|
|
2387
|
+
// No cache-write line item exists for any MiniMax model (billing is new
|
|
2388
|
+
// input + cache hit + output only — prompt-caching page, 2026-10-19), so
|
|
2389
|
+
// cache writes bill at the input rate.
|
|
2390
|
+
cacheWritePricePerMTok: 0.15,
|
|
2391
|
+
// No US re-host exists yet (open weights released 2026-10-19; not on
|
|
2392
|
+
// DeepInfra at verification time) — pinned to the native host like m2.5.
|
|
2393
|
+
regions: ['cn'],
|
|
2394
|
+
// Not published by MiniMax — M3's HF-template-verified cutoff carried
|
|
2395
|
+
// forward as the floor (this model shipped 4 months after M3).
|
|
2396
|
+
knowledgeCutoff: '2026-01-01',
|
|
2397
|
+
},
|
|
2331
2398
|
{
|
|
2332
2399
|
id: 'minimax-m3',
|
|
2333
2400
|
provider: 'minimax',
|
|
@@ -2396,9 +2463,12 @@ export const MODELS = [
|
|
|
2396
2463
|
us: { inputPricePerMTok: 0.25, outputPricePerMTok: 1, cacheReadPricePerMTok: 0.05 },
|
|
2397
2464
|
},
|
|
2398
2465
|
knowledgeCutoff: '2025-09-01',
|
|
2399
|
-
// Superseded by minimax-m3
|
|
2466
|
+
// Superseded by minimax-m3.1-flash — MiniMax's own successor for the M2.x
|
|
2467
|
+
// line (M/Token plan traffic to M2.5/M2.7 is auto-routed there). Closed to
|
|
2468
|
+
// new users 2026-10-19; retired upstream 2026-11-18. Kept priceable until
|
|
2469
|
+
// calls stop answering.
|
|
2400
2470
|
deprecatedAt: '2026-07-28',
|
|
2401
|
-
supersededBy: 'minimax-m3',
|
|
2471
|
+
supersededBy: 'minimax-m3.1-flash',
|
|
2402
2472
|
},
|
|
2403
2473
|
{
|
|
2404
2474
|
id: 'minimax-m2.5',
|
|
@@ -2421,9 +2491,11 @@ export const MODELS = [
|
|
|
2421
2491
|
// No US re-host exists (not on DeepInfra) — pinned to native China.
|
|
2422
2492
|
regions: ['cn'],
|
|
2423
2493
|
knowledgeCutoff: '2025-01-01',
|
|
2424
|
-
// Superseded by minimax-m3 (
|
|
2494
|
+
// Superseded by minimax-m3.1-flash (same hand-off as m2.7 — MiniMax routes
|
|
2495
|
+
// retired M2.x plan traffic there; closed to new users 2026-10-19, retired
|
|
2496
|
+
// upstream 2026-11-18). Kept priceable until calls stop answering.
|
|
2425
2497
|
deprecatedAt: '2026-03-18',
|
|
2426
|
-
supersededBy: 'minimax-m3',
|
|
2498
|
+
supersededBy: 'minimax-m3.1-flash',
|
|
2427
2499
|
},
|
|
2428
2500
|
// ---------------------------------------------------------------------------
|
|
2429
2501
|
// Alibaba (Qwen)
|
package/package.json
CHANGED