@molecule/api-resource-ai-models 1.17.3 → 1.18.1
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/README.md +6 -6
- package/dist/models.d.ts +11 -0
- package/dist/models.d.ts.map +1 -1
- package/dist/models.js +153 -17
- package/dist/types.d.ts +4 -5
- package/dist/types.d.ts.map +1 -1
- package/package.json +5 -5
package/README.md
CHANGED
|
@@ -3,7 +3,7 @@ AUTO-GENERATED — DO NOT EDIT THIS FILE.
|
|
|
3
3
|
Generated by `mlcl sync-docs` from the package's src/index.ts JSDoc + mlcl/registry.json.
|
|
4
4
|
Edits here are overwritten on the next commit (molecule's pre-commit hook regenerates).
|
|
5
5
|
To change this document, edit the module-level JSDoc in src/index.ts.
|
|
6
|
-
Generated: 2026-10-
|
|
6
|
+
Generated: 2026-10-22T22:44:54.731Z
|
|
7
7
|
-->
|
|
8
8
|
|
|
9
9
|
# @molecule/api-resource-ai-models
|
|
@@ -107,13 +107,12 @@ interface ModelDefinition {
|
|
|
107
107
|
* level names — e.g. Claude Haiku 4.5, Qwen3.7) list scaled-budget LABELS,
|
|
108
108
|
* e.g. `['4K', '8K', '16K', '32K']`, with {@link effortBudgetTokens} mapping
|
|
109
109
|
* each label to the actual token budget sent.
|
|
110
|
-
* - **
|
|
111
|
-
* `['off', 'low', 'high', 'max']` on DeepSeek V4.1 Flash
|
|
112
|
-
*
|
|
110
|
+
* - **Optional reasoning models** put the LOCAL `'off'` toggle first, e.g.
|
|
111
|
+
* `['off', 'low', 'high', 'max']` on DeepSeek V4.1 Flash. Their declared
|
|
112
|
+
* default may select reasoning. `'off'` is never sent to the provider: the
|
|
113
113
|
* resolved level produces no thinking block, which turns the vendor's
|
|
114
114
|
* reasoning off. A persisted value the model does not offer resolves to
|
|
115
|
-
*
|
|
116
|
-
* reasoning on by accident.
|
|
115
|
+
* its declared default rather than escalating through a cross-model rank.
|
|
117
116
|
* - **Fixed-reasoning models** (DeepSeek V4 Pro / V4 Flash, Kimi, …) omit this
|
|
118
117
|
* field entirely — reasoning depth can't be tuned, so there is nothing to
|
|
119
118
|
* pick.
|
|
@@ -221,6 +220,7 @@ interface ModelDefinition {
|
|
|
221
220
|
/**
|
|
222
221
|
* Regions in which this model is free-tier selectable even though the model
|
|
223
222
|
* as a whole is not `freeTier` — for models whose regional hosts price very
|
|
223
|
+
* differently (e.g. a cheap native host powering the free planner while its
|
|
224
224
|
// … truncated — the full declaration is in node_modules/@molecule/api-resource-ai-models/dist/types.d.ts
|
|
225
225
|
```
|
|
226
226
|
|
package/dist/models.d.ts
CHANGED
|
@@ -285,6 +285,17 @@ import type { ModelDefinition } from './types.js';
|
|
|
285
285
|
* labeled Kimi-K3 — so kimi-k2.7-code is native-only (regions ['cn']) and
|
|
286
286
|
* its US modelMap entry is gone. kimi-k3's US re-host is healthy, rates
|
|
287
287
|
* unchanged.)
|
|
288
|
+
* (re-verified 2026-10-23 across platform.kimi.ai — docs/models,
|
|
289
|
+
* docs/pricing/chat (server-rendered table), docs/api/models-overview and
|
|
290
|
+
* docs/guide/kimi-k3-1-pro-quickstart: K3.1 series released —
|
|
291
|
+
* kimi-k3.1-pro-highspeed added ($4/$20, cached input $0.20, 5m cache
|
|
292
|
+
* writes $4 / 1h $8; reasoning_effort low|high|max; max_completion_tokens
|
|
293
|
+
* default 131072 / max 1048576 — also re-read for kimi-k3, whose 65K
|
|
294
|
+
* "conservative" cap was stale and is now the documented 131072). kimi-k3
|
|
295
|
+
* superseded by kimi-k3.1-pro-highspeed. DeepInfra lists NO K3.1 re-host
|
|
296
|
+
* (only Kimi-K3 / Kimi-K2.6), so the new model is native-only. Banner:
|
|
297
|
+
* K2-series stops for new signups 2026-10-31, full retirement planned end
|
|
298
|
+
* of December 2026 — kimi-k2.6/k2.7-code/k2.5 need a retirement pass then.)
|
|
288
299
|
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
289
300
|
* (unchanged; minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
290
301
|
* (re-verified 2026-10-19 across platform.minimax.io — pricing-paygo,
|
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;AA2DjD
|
|
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;AA2DjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2YG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EAqlFnC,CAAA"}
|
package/dist/models.js
CHANGED
|
@@ -337,6 +337,17 @@ const CHINA_PUBLIC_HOLIDAYS = [
|
|
|
337
337
|
* labeled Kimi-K3 — so kimi-k2.7-code is native-only (regions ['cn']) and
|
|
338
338
|
* its US modelMap entry is gone. kimi-k3's US re-host is healthy, rates
|
|
339
339
|
* unchanged.)
|
|
340
|
+
* (re-verified 2026-10-23 across platform.kimi.ai — docs/models,
|
|
341
|
+
* docs/pricing/chat (server-rendered table), docs/api/models-overview and
|
|
342
|
+
* docs/guide/kimi-k3-1-pro-quickstart: K3.1 series released —
|
|
343
|
+
* kimi-k3.1-pro-highspeed added ($4/$20, cached input $0.20, 5m cache
|
|
344
|
+
* writes $4 / 1h $8; reasoning_effort low|high|max; max_completion_tokens
|
|
345
|
+
* default 131072 / max 1048576 — also re-read for kimi-k3, whose 65K
|
|
346
|
+
* "conservative" cap was stale and is now the documented 131072). kimi-k3
|
|
347
|
+
* superseded by kimi-k3.1-pro-highspeed. DeepInfra lists NO K3.1 re-host
|
|
348
|
+
* (only Kimi-K3 / Kimi-K2.6), so the new model is native-only. Banner:
|
|
349
|
+
* K2-series stops for new signups 2026-10-31, full retirement planned end
|
|
350
|
+
* of December 2026 — kimi-k2.6/k2.7-code/k2.5 need a retirement pass then.)
|
|
340
351
|
* - MiniMax: https://platform.minimax.io/docs/guides/pricing-paygo
|
|
341
352
|
* (unchanged; minimax-m3 $0.30/$1.20 is a "permanent 50% off" list rate)
|
|
342
353
|
* (re-verified 2026-10-19 across platform.minimax.io — pricing-paygo,
|
|
@@ -2120,12 +2131,9 @@ export const MODELS = [
|
|
|
2120
2131
|
description: 'Newest ultra-cheap & fast flash — vision-capable agentic coding',
|
|
2121
2132
|
contextWindow: 1_000_000,
|
|
2122
2133
|
maxOutputTokens: 384_000,
|
|
2123
|
-
//
|
|
2124
|
-
//
|
|
2125
|
-
//
|
|
2126
|
-
// non-thinking, so the catalog keeps that historical behavior: `off` is a
|
|
2127
|
-
// LOCAL toggle, never sent to the vendor as a reasoning_effort value. A
|
|
2128
|
-
// project that asks for `low`/`high`/`max` sends that native effort and the
|
|
2134
|
+
// Low is the default configurable effort. The explicit `off` selection is
|
|
2135
|
+
// a LOCAL toggle, never sent to the vendor as a reasoning_effort value. A
|
|
2136
|
+
// project using `low`/`high`/`max` sends that native effort and the
|
|
2129
2137
|
// bond replays every retained assistant reasoning_content through the tool
|
|
2130
2138
|
// loop (DeepSeek 400s a thinking-mode request with tools that omits one),
|
|
2131
2139
|
// which is what makes thinking mode usable on this surface at all.
|
|
@@ -2144,13 +2152,13 @@ export const MODELS = [
|
|
|
2144
2152
|
thinkingBudgetTokens: 8_000,
|
|
2145
2153
|
thinkingConfigurable: true,
|
|
2146
2154
|
supportedEffortLevels: ['off', 'low', 'high', 'max'],
|
|
2147
|
-
defaultEffortLevel: '
|
|
2155
|
+
defaultEffortLevel: 'low',
|
|
2148
2156
|
supportsVision: true,
|
|
2149
2157
|
supportsPromptCaching: true,
|
|
2150
2158
|
supportsTools: true,
|
|
2151
2159
|
// Free-tier default (moved from deepseek-v4-flash 2026-09-10): the
|
|
2152
2160
|
// cheapest capable executor on the go-forward id — fast, tool-calling,
|
|
2153
|
-
// vision-capable,
|
|
2161
|
+
// vision-capable, with low default effort for configurable chat. Serving cost
|
|
2154
2162
|
// is ~2× the retired US leg on output-heavy turns (CN 0.6/1.2 output vs
|
|
2155
2163
|
// DeepInfra's flat 0.18) — tiers.ts's free-turn budget math derives from
|
|
2156
2164
|
// this entry. Exactly one model in this catalog may carry freeTier
|
|
@@ -2198,8 +2206,23 @@ export const MODELS = [
|
|
|
2198
2206
|
},
|
|
2199
2207
|
// ---------------------------------------------------------------------------
|
|
2200
2208
|
// Moonshot (Kimi)
|
|
2201
|
-
// Verified: https://platform.kimi.ai/docs/
|
|
2202
|
-
// /docs/guide/
|
|
2209
|
+
// Verified: https://platform.kimi.ai/docs/pricing/chat +
|
|
2210
|
+
// /docs/api/models-overview + /docs/guide/kimi-k3-1-pro-quickstart
|
|
2211
|
+
// (2026-10-23, kimi-k3.1-pro add: exact id; $2/$10, cache read
|
|
2212
|
+
// $0.20, cache write $2 (5m, default TTL) / $4 (1h); 1M context;
|
|
2213
|
+
// max_completion_tokens default 131072 / settable 1048576;
|
|
2214
|
+
// text+image+video input; reasoning_effort low|high|max default
|
|
2215
|
+
// max; temperature fixed 1.0; tool_choice 'required' supported;
|
|
2216
|
+
// DeepInfra live-checked — no K3.1 re-host). Earlier:
|
|
2217
|
+
// /docs/models + /docs/guide/use-kimi-k2-thinking-model
|
|
2218
|
+
// (2026-07-28).
|
|
2219
|
+
// 2026-10-23: K3.1 series released (kimi-k3.1-pro + the -highspeed speed
|
|
2220
|
+
// tier, same intelligence at 2-3× the speed). Both variants are cataloged
|
|
2221
|
+
// (owner ✅, one card each): the provider's Model Parameter Reference states
|
|
2222
|
+
// the k3.1 pair has the SAME constraints as kimi-k3 — reasoning_effort
|
|
2223
|
+
// low|high|max (default max), temperature/top_p/n/penalties fixed, and
|
|
2224
|
+
// tool_choice 'required' supported. Native-only for now (no DeepInfra
|
|
2225
|
+
// re-host yet).
|
|
2203
2226
|
// The moonshot bond now supports PRESERVED THINKING (ChatMessage.reasoning →
|
|
2204
2227
|
// reasoning_content replay through tool loops), which unblocked the two
|
|
2205
2228
|
// previously-excluded models: kimi-k3 (2026-07-16 flagship — 2.8T MoE, 1M
|
|
@@ -2214,6 +2237,104 @@ export const MODELS = [
|
|
|
2214
2237
|
// region-re-host coverage check fails on exactly that (a us-region moonshot
|
|
2215
2238
|
// model missing from the bond's modelMap).
|
|
2216
2239
|
// ---------------------------------------------------------------------------
|
|
2240
|
+
{
|
|
2241
|
+
id: 'kimi-k3.1-pro',
|
|
2242
|
+
provider: 'moonshot',
|
|
2243
|
+
// Same parameter constraints as kimi-k3 (provider's own Model Parameter
|
|
2244
|
+
// Reference, api/models-overview, names the k3.1 pair): temperature is
|
|
2245
|
+
// fixed at 1.0 upstream — "passing any other value returns an error" — so
|
|
2246
|
+
// callers omit it. tool_choice 'required' IS supported (their FAQ names
|
|
2247
|
+
// kimi-k3.1-pro, kimi-k3.1-pro-highspeed and kimi-k3 as supporting it),
|
|
2248
|
+
// so no rejectsForcedToolChoice.
|
|
2249
|
+
rejectsTemperature: true,
|
|
2250
|
+
label: 'Kimi K3.1 Pro',
|
|
2251
|
+
description: "Moonshot's most capable flagship — same intelligence as the highspeed tier at half the price",
|
|
2252
|
+
contextWindow: 1_000_000,
|
|
2253
|
+
// Docs (guide/kimi-k3-1-pro-quickstart): max_completion_tokens defaults to
|
|
2254
|
+
// 131072 and is settable up to 1048576 (the full window). Cap at the
|
|
2255
|
+
// documented default so a request never reserves the whole window for
|
|
2256
|
+
// output and leaves room for a real prompt.
|
|
2257
|
+
maxOutputTokens: 131_072,
|
|
2258
|
+
supportsThinking: true,
|
|
2259
|
+
thinkingBudgetTokens: 8_000,
|
|
2260
|
+
// Thinking is FORCED ON (K3 family); depth rides reasoning_effort
|
|
2261
|
+
// low|high|max, upstream default max — we default to high like kimi-k3 so
|
|
2262
|
+
// agentic loops aren't pinned to the slowest/most expensive tier.
|
|
2263
|
+
thinkingConfigurable: true,
|
|
2264
|
+
supportedEffortLevels: ['low', 'high', 'max'],
|
|
2265
|
+
defaultEffortLevel: 'high',
|
|
2266
|
+
// Text + image (base64 / ms:// file id) + video input — the quickstart's
|
|
2267
|
+
// vision section; the highspeed tier shares this API surface.
|
|
2268
|
+
supportsVision: true,
|
|
2269
|
+
supportsPromptCaching: true,
|
|
2270
|
+
supportsTools: true,
|
|
2271
|
+
// Verified 2026-10-23 on platform.kimi.ai/docs/pricing/chat (K3 series
|
|
2272
|
+
// table): $2/$10, cached input $0.20. Cache writes are billed by TTL —
|
|
2273
|
+
// $2 (5m, the default tier when no TTL is set) / $4 (1h);
|
|
2274
|
+
// cacheWritePricePerMTok carries the default 5m tier, matching the
|
|
2275
|
+
// kimi-k3 / kimi-k3.1-pro-highspeed entries' convention.
|
|
2276
|
+
inputPricePerMTok: 2,
|
|
2277
|
+
outputPricePerMTok: 10,
|
|
2278
|
+
cacheReadPricePerMTok: 0.2,
|
|
2279
|
+
cacheWritePricePerMTok: 2,
|
|
2280
|
+
// Native-only: released 2026-10-23 and DeepInfra (live
|
|
2281
|
+
// api.deepinfra.com/v1/openai/models, checked same day) still lists only
|
|
2282
|
+
// Kimi-K3 and Kimi-K2.6 — no K3.1 re-host, so no regionPricing and no
|
|
2283
|
+
// bond modelMap entry (a us-region moonshot model missing from the map
|
|
2284
|
+
// would 404 at dispatch).
|
|
2285
|
+
regions: ['cn'],
|
|
2286
|
+
// Not published — best-effort estimate (mirrors the highspeed entry: same
|
|
2287
|
+
// weights generation, same intelligence).
|
|
2288
|
+
knowledgeCutoff: '2026-05-01',
|
|
2289
|
+
},
|
|
2290
|
+
{
|
|
2291
|
+
id: 'kimi-k3.1-pro-highspeed',
|
|
2292
|
+
provider: 'moonshot',
|
|
2293
|
+
// Same parameter constraints as kimi-k3 (provider's own Model Parameter
|
|
2294
|
+
// Reference, api/models-overview, names the k3.1 pair): temperature is
|
|
2295
|
+
// fixed at 1.0 upstream — "passing any other value returns an error" — so
|
|
2296
|
+
// callers omit it. tool_choice 'required' IS supported (their FAQ names
|
|
2297
|
+
// kimi-k3.1-pro, kimi-k3.1-pro-highspeed and kimi-k3 as supporting it),
|
|
2298
|
+
// so no rejectsForcedToolChoice.
|
|
2299
|
+
rejectsTemperature: true,
|
|
2300
|
+
label: 'Kimi K3.1 Pro Highspeed',
|
|
2301
|
+
description: 'Newest Moonshot flagship, speed tier — same intelligence as K3.1 Pro at 2-3× the speed',
|
|
2302
|
+
contextWindow: 1_000_000,
|
|
2303
|
+
// Docs (guide/kimi-k3-1-pro-quickstart): max_completion_tokens defaults to
|
|
2304
|
+
// 131072 and is settable up to 1048576 (the full window). Cap at the
|
|
2305
|
+
// documented default so a request never reserves the whole window for
|
|
2306
|
+
// output and leaves room for a real prompt.
|
|
2307
|
+
maxOutputTokens: 131_072,
|
|
2308
|
+
supportsThinking: true,
|
|
2309
|
+
thinkingBudgetTokens: 8_000,
|
|
2310
|
+
// Thinking is FORCED ON (K3 family); depth rides reasoning_effort
|
|
2311
|
+
// low|high|max, upstream default max — we default to high like kimi-k3 so
|
|
2312
|
+
// agentic loops aren't pinned to the slowest/most expensive tier.
|
|
2313
|
+
thinkingConfigurable: true,
|
|
2314
|
+
supportedEffortLevels: ['low', 'high', 'max'],
|
|
2315
|
+
defaultEffortLevel: 'high',
|
|
2316
|
+
// Text + image + video input — same vision API surface as kimi-k3.1-pro
|
|
2317
|
+
// (the only documented difference of the highspeed tier is speed).
|
|
2318
|
+
supportsVision: true,
|
|
2319
|
+
supportsPromptCaching: true,
|
|
2320
|
+
supportsTools: true,
|
|
2321
|
+
// Verified 2026-10-23 on platform.kimi.ai/docs/pricing/chat: $4/$20,
|
|
2322
|
+
// cached input $0.20. Cache writes are billed by TTL — $4 (5m, the
|
|
2323
|
+
// default tier when no TTL is set) / $8 (1h); cacheWritePricePerMTok
|
|
2324
|
+
// carries the default 5m tier, matching the kimi-k3 entry's convention.
|
|
2325
|
+
inputPricePerMTok: 4,
|
|
2326
|
+
outputPricePerMTok: 20,
|
|
2327
|
+
cacheReadPricePerMTok: 0.2,
|
|
2328
|
+
cacheWritePricePerMTok: 4,
|
|
2329
|
+
// Native-only: released 2026-10-23 and DeepInfra (live
|
|
2330
|
+
// api.deepinfra.com/v1/openai/models, checked same day) still lists only
|
|
2331
|
+
// Kimi-K3 and Kimi-K2.6 — no K3.1 re-host, so no regionPricing and no
|
|
2332
|
+
// bond modelMap entry (a us-region moonshot model missing from the map
|
|
2333
|
+
// would 404 at dispatch).
|
|
2334
|
+
regions: ['cn'],
|
|
2335
|
+
// Not published — best-effort estimate.
|
|
2336
|
+
knowledgeCutoff: '2026-05-01',
|
|
2337
|
+
},
|
|
2217
2338
|
{
|
|
2218
2339
|
id: 'kimi-k3',
|
|
2219
2340
|
provider: 'moonshot',
|
|
@@ -2223,8 +2344,10 @@ export const MODELS = [
|
|
|
2223
2344
|
label: 'Kimi K3',
|
|
2224
2345
|
description: 'Moonshot flagship — 2.8T open weights, 1M context, multimodal',
|
|
2225
2346
|
contextWindow: 1_000_000,
|
|
2226
|
-
//
|
|
2227
|
-
|
|
2347
|
+
// Docs (guide/kimi-k3-quickstart, re-read 2026-10-23): max_completion_tokens
|
|
2348
|
+
// defaults to 131072, settable up to 1048576 — the old "not documented"
|
|
2349
|
+
// 65K conservative cap was stale. Same cap as the k3.1 entries.
|
|
2350
|
+
maxOutputTokens: 131_072,
|
|
2228
2351
|
supportsThinking: true,
|
|
2229
2352
|
thinkingBudgetTokens: 8_000,
|
|
2230
2353
|
// Thinking is FORCED ON; depth rides reasoning_effort (low|high|max). The
|
|
@@ -2261,6 +2384,14 @@ export const MODELS = [
|
|
|
2261
2384
|
},
|
|
2262
2385
|
// Not published — best-effort estimate.
|
|
2263
2386
|
knowledgeCutoff: '2026-01-01',
|
|
2387
|
+
// Superseded by kimi-k3.1-pro (2026-10-23): the K3.1 series is the new
|
|
2388
|
+
// flagship generation — the provider's own models page calls kimi-k3 the
|
|
2389
|
+
// "prior flagship" and kimi-k3.1-pro "Kimi's most capable flagship to
|
|
2390
|
+
// date" (same intelligence as the highspeed tier at half the price, so the
|
|
2391
|
+
// base tier is the true successor of the old flagship). Kept priceable
|
|
2392
|
+
// and dispatchable on both regions.
|
|
2393
|
+
deprecatedAt: '2026-10-23',
|
|
2394
|
+
supersededBy: 'kimi-k3.1-pro',
|
|
2264
2395
|
},
|
|
2265
2396
|
{
|
|
2266
2397
|
id: 'kimi-k2.7-code',
|
|
@@ -2310,6 +2441,11 @@ export const MODELS = [
|
|
|
2310
2441
|
{
|
|
2311
2442
|
id: 'kimi-k2.6',
|
|
2312
2443
|
provider: 'moonshot',
|
|
2444
|
+
// Native Moonshot host moved K2.6 to fixed temperatures — the parameter
|
|
2445
|
+
// reference says "1.0 thinking / 0.6 non-thinking", and a live
|
|
2446
|
+
// 2026-10-23 probe of the plain shape (temperature 0) got 400 "invalid
|
|
2447
|
+
// temperature: only 0.6 is allowed for this model". Callers omit it.
|
|
2448
|
+
rejectsTemperature: true,
|
|
2313
2449
|
label: 'Kimi K2.6',
|
|
2314
2450
|
description: 'Multimodal agent — strong sequential tool use',
|
|
2315
2451
|
contextWindow: 262_144,
|
|
@@ -2336,9 +2472,9 @@ export const MODELS = [
|
|
|
2336
2472
|
us: { inputPricePerMTok: 0.75, outputPricePerMTok: 3.5, cacheReadPricePerMTok: 0.15 },
|
|
2337
2473
|
},
|
|
2338
2474
|
knowledgeCutoff: '2025-04-01',
|
|
2339
|
-
// Superseded by
|
|
2475
|
+
// Superseded by the current selectable Kimi flagship — kept priceable.
|
|
2340
2476
|
deprecatedAt: '2026-07-28',
|
|
2341
|
-
supersededBy: 'kimi-k3',
|
|
2477
|
+
supersededBy: 'kimi-k3.1-pro',
|
|
2342
2478
|
},
|
|
2343
2479
|
{
|
|
2344
2480
|
id: 'kimi-k2.5',
|
|
@@ -2366,8 +2502,8 @@ export const MODELS = [
|
|
|
2366
2502
|
},
|
|
2367
2503
|
knowledgeCutoff: '2024-04-01',
|
|
2368
2504
|
// Two generations behind. `supersededBy` names the current selectable Kimi
|
|
2369
|
-
//
|
|
2370
|
-
// forward in one hop. Moonshot RETIRED kimi-k2.5 on 2026-08-31 —
|
|
2505
|
+
// flagship rather than the also-superseded k2.6/k3, so a saved selection
|
|
2506
|
+
// resolves forward in one hop. Moonshot RETIRED kimi-k2.5 on 2026-08-31 —
|
|
2371
2507
|
// platform.kimi.ai/docs/models says calls 404, and a live 2026-09-10
|
|
2372
2508
|
// dispatch probe confirmed the native host refusing it. The DeepInfra
|
|
2373
2509
|
// re-host still answered inference on 2026-09-10 but is DELISTED from its
|
|
@@ -2376,7 +2512,7 @@ export const MODELS = [
|
|
|
2376
2512
|
// selections still resolve forward.
|
|
2377
2513
|
deprecatedAt: '2026-04-01',
|
|
2378
2514
|
disabled: true,
|
|
2379
|
-
supersededBy: 'kimi-k3',
|
|
2515
|
+
supersededBy: 'kimi-k3.1-pro',
|
|
2380
2516
|
},
|
|
2381
2517
|
// ---------------------------------------------------------------------------
|
|
2382
2518
|
// MiniMax
|
package/dist/types.d.ts
CHANGED
|
@@ -80,13 +80,12 @@ export interface ModelDefinition {
|
|
|
80
80
|
* level names — e.g. Claude Haiku 4.5, Qwen3.7) list scaled-budget LABELS,
|
|
81
81
|
* e.g. `['4K', '8K', '16K', '32K']`, with {@link effortBudgetTokens} mapping
|
|
82
82
|
* each label to the actual token budget sent.
|
|
83
|
-
* - **
|
|
84
|
-
* `['off', 'low', 'high', 'max']` on DeepSeek V4.1 Flash
|
|
85
|
-
*
|
|
83
|
+
* - **Optional reasoning models** put the LOCAL `'off'` toggle first, e.g.
|
|
84
|
+
* `['off', 'low', 'high', 'max']` on DeepSeek V4.1 Flash. Their declared
|
|
85
|
+
* default may select reasoning. `'off'` is never sent to the provider: the
|
|
86
86
|
* resolved level produces no thinking block, which turns the vendor's
|
|
87
87
|
* reasoning off. A persisted value the model does not offer resolves to
|
|
88
|
-
*
|
|
89
|
-
* reasoning on by accident.
|
|
88
|
+
* its declared default rather than escalating through a cross-model rank.
|
|
90
89
|
* - **Fixed-reasoning models** (DeepSeek V4 Pro / V4 Flash, Kimi, …) omit this
|
|
91
90
|
* field entirely — reasoning depth can't be tuned, so there is nothing to
|
|
92
91
|
* pick.
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH;;;;;GAKG;AACH,MAAM,MAAM,YAAY,GACpB,WAAW,GACX,QAAQ,GACR,QAAQ,GACR,KAAK,GACL,UAAU,GACV,MAAM,GACN,UAAU,GACV,SAAS,GACT,SAAS,GACT,OAAO;AACT;;;;;GAKG;GACD,QAAQ,CAAA;AAEZ;;;;;;;;;;;;;;;GAeG;AACH,MAAM,MAAM,WAAW,GAAG,MAAM,CAAA;AAEhC;;;GAGG;AACH,MAAM,WAAW,eAAe;IAC9B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAA;IACV,2CAA2C;IAC3C,QAAQ,EAAE,YAAY,CAAA;IACtB,yDAAyD;IACzD,KAAK,EAAE,MAAM,CAAA;IACb,wCAAwC;IACxC,WAAW,EAAE,MAAM,CAAA;IACnB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,CAAA;IACrB,0CAA0C;IAC1C,eAAe,EAAE,MAAM,CAAA;IACvB,uEAAuE;IACvE,gBAAgB,EAAE,OAAO,CAAA;IACzB,yFAAyF;IACzF,oBAAoB,EAAE,MAAM,CAAA;IAC5B;;;OAGG;IACH,oBAAoB,EAAE,OAAO,CAAA;IAC7B
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH;;;;;GAKG;AACH,MAAM,MAAM,YAAY,GACpB,WAAW,GACX,QAAQ,GACR,QAAQ,GACR,KAAK,GACL,UAAU,GACV,MAAM,GACN,UAAU,GACV,SAAS,GACT,SAAS,GACT,OAAO;AACT;;;;;GAKG;GACD,QAAQ,CAAA;AAEZ;;;;;;;;;;;;;;;GAeG;AACH,MAAM,MAAM,WAAW,GAAG,MAAM,CAAA;AAEhC;;;GAGG;AACH,MAAM,WAAW,eAAe;IAC9B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAA;IACV,2CAA2C;IAC3C,QAAQ,EAAE,YAAY,CAAA;IACtB,yDAAyD;IACzD,KAAK,EAAE,MAAM,CAAA;IACb,wCAAwC;IACxC,WAAW,EAAE,MAAM,CAAA;IACnB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,CAAA;IACrB,0CAA0C;IAC1C,eAAe,EAAE,MAAM,CAAA;IACvB,uEAAuE;IACvE,gBAAgB,EAAE,OAAO,CAAA;IACzB,yFAAyF;IACzF,oBAAoB,EAAE,MAAM,CAAA;IAC5B;;;OAGG;IACH,oBAAoB,EAAE,OAAO,CAAA;IAC7B;;;;;;;;;;;;;;;;;;;;;;;;;;OA0BG;IACH,qBAAqB,CAAC,EAAE,WAAW,EAAE,CAAA;IACrC;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,WAAW,CAAA;IAChC;;;;;;;;;;;OAWG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC3C,mEAAmE;IACnE,cAAc,EAAE,OAAO,CAAA;IACvB,iDAAiD;IACjD,qBAAqB,EAAE,OAAO,CAAA;IAC9B,8DAA8D;IAC9D,aAAa,EAAE,OAAO,CAAA;IACtB;;;;;;;;;;;;;OAaG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAA;IAClC;;;;;;;;;;;OAWG;IACH,uBAAuB,CAAC,EAAE,OAAO,CAAA;IACjC;;;;;;;;;;OAUG;IACH,kBAAkB,CAAC,EAAE,OAAO,CAAA;IAC5B;;;;;;;;OAQG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B;;;;;;;OAOG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B;;;OAGG;IACH,qBAAqB,CAAC,EAAE,MAAM,CAAA;IAC9B;;;OAGG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,wFAAwF;IACxF,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,MAAM,EAAE,CAAA;IAC1B;;;;;;;OAOG;IACH,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;IAClB;;;;;;;;OAQG;IACH,aAAa,CAAC,EAAE,MAAM,CACpB,MAAM,EACN;QACE,6DAA6D;QAC7D,iBAAiB,EAAE,MAAM,CAAA;QACzB,qDAAqD;QACrD,kBAAkB,EAAE,MAAM,CAAA;QAC1B,gEAAgE;QAChE,qBAAqB,CAAC,EAAE,MAAM,CAAA;QAC9B,iEAAiE;QACjE,sBAAsB,CAAC,EAAE,MAAM,CAAA;KAChC,CACF,CAAA;IACD,sEAAsE;IACtE,iBAAiB,EAAE,MAAM,CAAA;IACzB,8CAA8C;IAC9C,kBAAkB,EAAE,MAAM,CAAA;IAC1B;;;;;;;;;;;OAWG;IACH,qBAAqB,EAAE,MAAM,CAAA;IAC7B;;;;;;;;;;OAUG;IACH,sBAAsB,EAAE,MAAM,CAAA;IAC9B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OAkCG;IACH,WAAW,CAAC,EAAE;QACZ,OAAO,EAAE;YAAE,cAAc,EAAE,MAAM,CAAC;YAAC,YAAY,EAAE,MAAM,CAAC;YAAC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;SAAE,EAAE,CAAA;QACrF,UAAU,EAAE,MAAM,CAAA;QAClB,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;QAC3B,IAAI,CAAC,EAAE;YAAE,GAAG,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KACrC,CAAA;IACD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OAwCG;IACH,gBAAgB,CAAC,EAAE;QACjB,sDAAsD;QACtD,aAAa,EAAE,MAAM,CAAA;QACrB,8EAA8E;QAC9E,iBAAiB,EAAE,MAAM,CAAA;QACzB,oEAAoE;QACpE,kBAAkB,EAAE,MAAM,CAAA;QAC1B,iFAAiF;QACjF,qBAAqB,EAAE,MAAM,CAAA;QAC7B,kFAAkF;QAClF,sBAAsB,EAAE,MAAM,CAAA;QAC9B,yFAAyF;QACzF,WAAW,CAAC,EAAE;YACZ,OAAO,EAAE;gBAAE,cAAc,EAAE,MAAM,CAAC;gBAAC,YAAY,EAAE,MAAM,CAAC;gBAAC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;aAAE,EAAE,CAAA;YACrF,UAAU,EAAE,MAAM,CAAA;YAClB,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;YAC3B,IAAI,CAAC,EAAE;gBAAE,GAAG,EAAE,MAAM,CAAC;gBAAC,IAAI,EAAE,MAAM,CAAA;aAAE,CAAA;SACrC,CAAA;QACD;;;;WAIG;QACH,kBAAkB,CAAC,EAAE,eAAe,CAAC,oBAAoB,CAAC,CAAA;QAC1D,6EAA6E;QAC7E,MAAM,CAAC,EAAE,MAAM,CAAA;KAChB,CAAA;IACD;;;;;;;;;;OAUG;IACH,WAAW,CAAC,EAAE;QACZ,gEAAgE;QAChE,iBAAiB,EAAE,MAAM,CAAA;QACzB,wDAAwD;QACxD,kBAAkB,EAAE,MAAM,CAAA;QAC1B,mEAAmE;QACnE,qBAAqB,EAAE,MAAM,CAAA;QAC7B,oEAAoE;QACpE,sBAAsB,EAAE,MAAM,CAAA;KAC/B,CAAA;IACD;;;;;;;;OAQG;IACH,kBAAkB,CAAC,EAAE;QACnB,6EAA6E;QAC7E,gBAAgB,EAAE,MAAM,CAAA;QACxB,mEAAmE;QACnE,iBAAiB,EAAE,MAAM,CAAA;QACzB,2DAA2D;QAC3D,kBAAkB,EAAE,MAAM,CAAA;QAC1B,sEAAsE;QACtE,qBAAqB,EAAE,MAAM,CAAA;QAC7B,uEAAuE;QACvE,sBAAsB,EAAE,MAAM,CAAA;KAC/B,CAAA;IACD,mDAAmD;IACnD,eAAe,EAAE,MAAM,CAAA;IACvB;;;;;;;;;OASG;IACH,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB;;;;;;;;;;;;;;OAcG;IACH,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB;;;;;;;;;;;;;;;;;;;;;;;;;;;OA2BG;IACH,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB;AAED;;;;;;GAMG;AACH,MAAM,WAAW,iBAAiB;IAChC,6DAA6D;IAC7D,IAAI,EAAE,MAAM,CAAA;IACZ,gEAAgE;IAChE,OAAO,EAAE,MAAM,CAAA;IACf,8EAA8E;IAC9E,MAAM,EAAE,MAAM,CAAA;IACd,4EAA4E;IAC5E,OAAO,EAAE,MAAM,CAAA;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,eAAe,EAAE,CAAA;IACzB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,iBAAiB,CAAA;CAC7B"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@molecule/api-resource-ai-models",
|
|
3
|
-
"version": "1.
|
|
3
|
+
"version": "1.18.1",
|
|
4
4
|
"description": "AI model catalog — server-side source of truth plus an authentication-gated discovery endpoint",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "dist/index.js",
|
|
@@ -27,10 +27,10 @@
|
|
|
27
27
|
],
|
|
28
28
|
"license": "Apache-2.0",
|
|
29
29
|
"devDependencies": {
|
|
30
|
-
"@molecule/api-bond": "1.0.
|
|
31
|
-
"@molecule/api-i18n": "1.0.
|
|
32
|
-
"@molecule/api-resource": "1.0.
|
|
33
|
-
"@types/node": "26.
|
|
30
|
+
"@molecule/api-bond": "1.0.3",
|
|
31
|
+
"@molecule/api-i18n": "1.0.4",
|
|
32
|
+
"@molecule/api-resource": "1.0.6",
|
|
33
|
+
"@types/node": "26.6.5",
|
|
34
34
|
"typescript": "6.0.3",
|
|
35
35
|
"vitest": "4.1.12"
|
|
36
36
|
},
|