@molecule/api-resource-ai-models 1.2.7 → 1.4.0
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 +62 -9
- package/dist/lookup.d.ts +2 -1
- package/dist/lookup.d.ts.map +1 -1
- package/dist/lookup.js +17 -6
- package/dist/models.d.ts +41 -4
- package/dist/models.d.ts.map +1 -1
- package/dist/models.js +172 -33
- package/dist/types.d.ts +12 -0
- package/dist/types.d.ts.map +1 -1
- package/package.json +1 -1
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-
|
|
6
|
+
Generated: 2026-09-01T19:40:01.931Z
|
|
7
7
|
-->
|
|
8
8
|
|
|
9
9
|
# @molecule/api-resource-ai-models
|
|
@@ -255,9 +255,19 @@ interface ModelDefinition {
|
|
|
255
255
|
*
|
|
256
256
|
* Windows are minutes-since-midnight UTC, half-open `[start, end)`; a window
|
|
257
257
|
* may wrap midnight (`start > end`).
|
|
258
|
+
*
|
|
259
|
+
* `daysOfWeekUtc` restricts a window to certain UTC weekdays (`0` = Sunday …
|
|
260
|
+
* `6` = Saturday); omitted means every day. A provider that prices its peak
|
|
261
|
+
* by BUSINESS hours qualifies them by day, and a day-less window over-bills
|
|
262
|
+
* every weekend hour inside it at the full multiplier — DeepSeek's card
|
|
263
|
+
* carried no day qualifier when these windows landed on 2026-08-16 and added
|
|
264
|
+
* "Monday through Friday" by 2026-08-31, which is exactly the 2× weekend
|
|
265
|
+
* over-bill this field exists to prevent. A wrapping window belongs to the
|
|
266
|
+
* day it STARTS on, so its post-midnight tail is still matched against the
|
|
267
|
+
* previous day.
|
|
258
268
|
*/
|
|
259
269
|
peakPricing?: {
|
|
260
|
-
windows: { startMinuteUtc: number; endMinuteUtc: number }[]
|
|
270
|
+
windows: { startMinuteUtc: number; endMinuteUtc: number; daysOfWeekUtc?: number[] }[]
|
|
261
271
|
multiplier: number
|
|
262
272
|
}
|
|
263
273
|
/**
|
|
@@ -305,7 +315,7 @@ interface ModelDefinition {
|
|
|
305
315
|
cacheWritePricePerMTok: number
|
|
306
316
|
/** Peak-hour pricing from `effectiveFrom` (omitted → existing windows carry through). */
|
|
307
317
|
peakPricing?: {
|
|
308
|
-
windows: { startMinuteUtc: number; endMinuteUtc: number }[]
|
|
318
|
+
windows: { startMinuteUtc: number; endMinuteUtc: number; daysOfWeekUtc?: number[] }[]
|
|
309
319
|
multiplier: number
|
|
310
320
|
}
|
|
311
321
|
/** Where the change was announced, for the re-verify pass after it lands. */
|
|
@@ -529,7 +539,12 @@ leaves the existing windows in force), else the model's own `peakPricing`.
|
|
|
529
539
|
function effectivePeakPricing(
|
|
530
540
|
modelDef: ModelDefinition,
|
|
531
541
|
at?: Date,
|
|
532
|
-
):
|
|
542
|
+
):
|
|
543
|
+
| {
|
|
544
|
+
windows: { startMinuteUtc: number; endMinuteUtc: number; daysOfWeekUtc?: number[] }[]
|
|
545
|
+
multiplier: number
|
|
546
|
+
}
|
|
547
|
+
| undefined
|
|
533
548
|
```
|
|
534
549
|
|
|
535
550
|
- `modelDef` — The model definition.
|
|
@@ -647,7 +662,8 @@ function modelRegionRates(modelDef: ModelDefinition, requested?: string, at?: Da
|
|
|
647
662
|
The price multiplier in effect for a model at a given instant, in a region.
|
|
648
663
|
|
|
649
664
|
Consults the model's {@link ModelDefinition.peakPricing} windows (UTC,
|
|
650
|
-
half-open, may wrap midnight
|
|
665
|
+
half-open, may wrap midnight, optionally restricted to certain UTC weekdays
|
|
666
|
+
via `daysOfWeekUtc`). Metering MUST call this with each request's
|
|
651
667
|
own timestamp so peak-hour usage bills at the provider's real rate — pricing
|
|
652
668
|
everything at the flat rate silently under-meters peak traffic.
|
|
653
669
|
|
|
@@ -777,6 +793,29 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
|
|
|
777
793
|
(re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
|
|
778
794
|
cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
|
|
779
795
|
models.dev first listed it 2026-08-06)
|
|
796
|
+
(re-verified 2026-09-01 on /docs/en/about-claude/pricing: sonnet-5's
|
|
797
|
+
$2/$10 launch pricing is now the STANDARD price — that page's own note
|
|
798
|
+
reads "The previously scheduled increase to $3/$15 per million input/output
|
|
799
|
+
tokens on September 1, 2026 will not occur." Catalog moved off the
|
|
800
|
+
never-under-charge $3/$15 to $2/$10, cache $0.20/$2.50, and the
|
|
801
|
+
KNOWN_DIVERGENCES entry was deleted. Every other Anthropic entry on that
|
|
802
|
+
page is unchanged: fable-5 $10/$50 ($1/$12.50), opus-5 / opus-4-8 /
|
|
803
|
+
opus-4-7 / opus-4-6 $5/$25 ($0.50/$6.25), sonnet-4-6 $3/$15 ($0.30/$3.75),
|
|
804
|
+
haiku-4-5 $1/$5 ($0.10/$1.25).)
|
|
805
|
+
(re-verified 2026-09-01 — ADDED claude-fable-5-1, released that day and
|
|
806
|
+
listed as "Active (latest)" on /docs/en/models/fable-5-1/overview, which
|
|
807
|
+
moves claude-fable-5 to "Legacy models (still available)" → superseded.
|
|
808
|
+
Same tier and same $10/$50 as fable-5, 1M ctx / 128K out, adaptive
|
|
809
|
+
thinking always on, effort low|medium|high|xhigh|max default high, vision
|
|
810
|
+
- tools + caching, reliable knowledge cutoff Jun 2026. The one price that
|
|
811
|
+
is NOT the fleet ratio: cache read is $0.25/MTok — the pricing page
|
|
812
|
+
footnotes fable-5-1 / mythos-5-1 as 0.025× input where every other model
|
|
813
|
+
is 0.1×; cache write is the usual 1.25× ($12.50). Not modeled: the beta
|
|
814
|
+
per-message effort / turn-scoped system messages / `display: "updates"`,
|
|
815
|
+
and the three fable-5 → fable-5-1 breaking changes (forced tool_choice
|
|
816
|
+
any/tool now 400s, thinking blocks are model-bound, editing earlier turns
|
|
817
|
+
invalidates them) — Synthase sends tool_choice auto and is append-only,
|
|
818
|
+
and the pre-commit dispatch probe covers the entry as sent.)
|
|
780
819
|
- OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
|
|
781
820
|
2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
|
|
782
821
|
20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
|
|
@@ -813,12 +852,16 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
|
|
|
813
852
|
grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
814
853
|
grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
815
854
|
- DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
816
|
-
2026-08-
|
|
855
|
+
2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
817
856
|
never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
818
857
|
has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
819
|
-
the same card introduced
|
|
820
|
-
|
|
821
|
-
|
|
858
|
+
the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
859
|
+
unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
860
|
+
06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
861
|
+
where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
862
|
+
`daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
863
|
+
also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
864
|
+
deliberately not catalogued.)
|
|
822
865
|
- Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
823
866
|
the US re-host (kimi-k3 flagship 2026-07-16
|
|
824
867
|
— 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
|
@@ -851,6 +894,16 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
|
|
|
851
894
|
$1.65/$4.951), so nothing would ever select it — and carrying both would
|
|
852
895
|
put two selectable Alibaba flagships in one family. Revisit only if Alibaba
|
|
853
896
|
publishes it as a distinct first-party DashScope model id.)
|
|
897
|
+
(re-verified 2026-08-28 on the per-model pages, which publish USD rates per
|
|
898
|
+
region directly — a better source than the CNY pricing table: qwen3.8-max's
|
|
899
|
+
Singapore implicit-cache rate is $0.25, NOT the 20%/$0.40 the ZH cache doc's
|
|
900
|
+
standard table implies. That doc names qwen3.8-max, qwen3.8-flash and
|
|
901
|
+
qwen3.8-2.4t-a95b as exceptions and defers to the console; the model pages
|
|
902
|
+
are the console's figures. qwen3.7-max's $0.50 (20%) is confirmed unchanged.
|
|
903
|
+
qwen3.8-flash (2026-08-26) ADDED: $0.15/$0.47, implicit cache $0.016, 1M ctx
|
|
904
|
+
/ 131,072 out, Image+Text+Video input, tools, hybrid thinking ON by default
|
|
905
|
+
with thinking_budget. It is cn-region-only — api.deepinfra.com has no
|
|
906
|
+
Qwen/Qwen3.8-Flash — so ALIBABA_US_MODEL_MAP needs no entry.)
|
|
854
907
|
- Zhipu: https://docs.z.ai/guides/overview/pricing + docs.z.ai/guides/llm/
|
|
855
908
|
glm-5.3 (verified 2026-08-26: glm-5.3 shipped 2026-08-14 and IS in the
|
|
856
909
|
catalog — $1.40/$4.40, cached $0.26, i.e. glm-5.2's card unchanged, 1M ctx,
|
package/dist/lookup.d.ts
CHANGED
|
@@ -111,7 +111,8 @@ export declare function withEffectivePricing(modelDef: ModelDefinition, at?: Dat
|
|
|
111
111
|
* The price multiplier in effect for a model at a given instant, in a region.
|
|
112
112
|
*
|
|
113
113
|
* Consults the model's {@link ModelDefinition.peakPricing} windows (UTC,
|
|
114
|
-
* half-open, may wrap midnight
|
|
114
|
+
* half-open, may wrap midnight, optionally restricted to certain UTC weekdays
|
|
115
|
+
* via `daysOfWeekUtc`). Metering MUST call this with each request's
|
|
115
116
|
* own timestamp so peak-hour usage bills at the provider's real rate — pricing
|
|
116
117
|
* everything at the flat rate silently under-meters peak traffic.
|
|
117
118
|
*
|
package/dist/lookup.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"lookup.d.ts","sourceRoot":"","sources":["../src/lookup.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,YAAY,CAAA;AAE/D;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAC/B,KAAK,EAAE,IAAI,CAAC,eAAe,EAAE,UAAU,GAAG,cAAc,CAAC,GACxD,OAAO,CAET;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,wBAAwB,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CASvE;AAED;;;;;;;GAOG;AACH,eAAO,MAAM,SAAS,EAAE,WAAW,CAAC,MAAM,CAEzC,CAAA;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAEhE;AAED;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,YAAY,GAAG,SAAS,eAAe,EAAE,CAEtF;AAED;;;;;;;;;GASG;AACH,wBAAgB,kBAAkB,CAChC,kBAAkB,EAAE,WAAW,CAAC,YAAY,CAAC,GAAG,SAAS,YAAY,EAAE,GACtE,SAAS,eAAe,EAAE,CAI5B;AAoBD;;;;;;;;;;;;GAYG;AACH,wBAAgB,kBAAkB,CAChC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CAgBjB;AAED;;;;;;;;;GASG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CAAC,aAAa,CAAC,CAMhC;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CASjB;AAED
|
|
1
|
+
{"version":3,"file":"lookup.d.ts","sourceRoot":"","sources":["../src/lookup.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,YAAY,CAAA;AAE/D;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAC/B,KAAK,EAAE,IAAI,CAAC,eAAe,EAAE,UAAU,GAAG,cAAc,CAAC,GACxD,OAAO,CAET;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,wBAAwB,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CASvE;AAED;;;;;;;GAOG;AACH,eAAO,MAAM,SAAS,EAAE,WAAW,CAAC,MAAM,CAEzC,CAAA;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAEhE;AAED;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,YAAY,GAAG,SAAS,eAAe,EAAE,CAEtF;AAED;;;;;;;;;GASG;AACH,wBAAgB,kBAAkB,CAChC,kBAAkB,EAAE,WAAW,CAAC,YAAY,CAAC,GAAG,SAAS,YAAY,EAAE,GACtE,SAAS,eAAe,EAAE,CAI5B;AAoBD;;;;;;;;;;;;GAYG;AACH,wBAAgB,kBAAkB,CAChC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CAgBjB;AAED;;;;;;;;;GASG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CAAC,aAAa,CAAC,CAMhC;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,eAAe,EACzB,EAAE,GAAE,IAAiB,GACpB,eAAe,CASjB;AAED;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAgB,iBAAiB,CAC/B,QAAQ,EAAE,eAAe,GAAG,SAAS,EACrC,EAAE,EAAE,IAAI,EACR,MAAM,CAAC,EAAE,MAAM,GACd,MAAM,CAsBR;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,eAAe,GAAG,SAAS,EACrC,SAAS,CAAC,EAAE,MAAM,GACjB,MAAM,CAGR;AAED,2DAA2D;AAC3D,MAAM,WAAW,eAAe;IAC9B,sDAAsD;IACtD,iBAAiB,EAAE,MAAM,CAAA;IACzB,8CAA8C;IAC9C,kBAAkB,EAAE,MAAM,CAAA;IAC1B,yDAAyD;IACzD,qBAAqB,EAAE,MAAM,CAAA;IAC7B,0DAA0D;IAC1D,sBAAsB,EAAE,MAAM,CAAA;CAC/B;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAgB,gBAAgB,CAC9B,QAAQ,EAAE,eAAe,EACzB,SAAS,CAAC,EAAE,MAAM,EAClB,EAAE,GAAE,IAAiB,GACpB,eAAe,CAYjB"}
|
package/dist/lookup.js
CHANGED
|
@@ -181,7 +181,8 @@ export function withEffectivePricing(modelDef, at = new Date()) {
|
|
|
181
181
|
* The price multiplier in effect for a model at a given instant, in a region.
|
|
182
182
|
*
|
|
183
183
|
* Consults the model's {@link ModelDefinition.peakPricing} windows (UTC,
|
|
184
|
-
* half-open, may wrap midnight
|
|
184
|
+
* half-open, may wrap midnight, optionally restricted to certain UTC weekdays
|
|
185
|
+
* via `daysOfWeekUtc`). Metering MUST call this with each request's
|
|
185
186
|
* own timestamp so peak-hour usage bills at the provider's real rate — pricing
|
|
186
187
|
* everything at the flat rate silently under-meters peak traffic.
|
|
187
188
|
*
|
|
@@ -210,11 +211,21 @@ export function priceMultiplierAt(modelDef, at, region) {
|
|
|
210
211
|
return 1;
|
|
211
212
|
const minute = at.getUTCHours() * 60 + at.getUTCMinutes();
|
|
212
213
|
for (const w of peak.windows) {
|
|
213
|
-
const
|
|
214
|
-
|
|
215
|
-
|
|
216
|
-
|
|
217
|
-
|
|
214
|
+
const wraps = w.startMinuteUtc > w.endMinuteUtc;
|
|
215
|
+
const inWindow = wraps
|
|
216
|
+
? minute >= w.startMinuteUtc || minute < w.endMinuteUtc
|
|
217
|
+
: minute >= w.startMinuteUtc && minute < w.endMinuteUtc;
|
|
218
|
+
if (!inWindow)
|
|
219
|
+
continue;
|
|
220
|
+
if (w.daysOfWeekUtc && w.daysOfWeekUtc.length > 0) {
|
|
221
|
+
// A wrapping window belongs to the day it STARTED on, so its
|
|
222
|
+
// post-midnight tail is matched against the previous UTC day — a
|
|
223
|
+
// Friday-only 23:00-02:00 window must still be peak at Saturday 00:30.
|
|
224
|
+
const day = wraps && minute < w.endMinuteUtc ? (at.getUTCDay() + 6) % 7 : at.getUTCDay();
|
|
225
|
+
if (!w.daysOfWeekUtc.includes(day))
|
|
226
|
+
continue;
|
|
227
|
+
}
|
|
228
|
+
return peak.multiplier;
|
|
218
229
|
}
|
|
219
230
|
return 1;
|
|
220
231
|
}
|
package/dist/models.d.ts
CHANGED
|
@@ -61,6 +61,29 @@ import type { ModelDefinition } from './types.js';
|
|
|
61
61
|
* (re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
|
|
62
62
|
* cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
|
|
63
63
|
* models.dev first listed it 2026-08-06)
|
|
64
|
+
* (re-verified 2026-09-01 on /docs/en/about-claude/pricing: sonnet-5's
|
|
65
|
+
* $2/$10 launch pricing is now the STANDARD price — that page's own note
|
|
66
|
+
* reads "The previously scheduled increase to $3/$15 per million input/output
|
|
67
|
+
* tokens on September 1, 2026 will not occur." Catalog moved off the
|
|
68
|
+
* never-under-charge $3/$15 to $2/$10, cache $0.20/$2.50, and the
|
|
69
|
+
* KNOWN_DIVERGENCES entry was deleted. Every other Anthropic entry on that
|
|
70
|
+
* page is unchanged: fable-5 $10/$50 ($1/$12.50), opus-5 / opus-4-8 /
|
|
71
|
+
* opus-4-7 / opus-4-6 $5/$25 ($0.50/$6.25), sonnet-4-6 $3/$15 ($0.30/$3.75),
|
|
72
|
+
* haiku-4-5 $1/$5 ($0.10/$1.25).)
|
|
73
|
+
* (re-verified 2026-09-01 — ADDED claude-fable-5-1, released that day and
|
|
74
|
+
* listed as "Active (latest)" on /docs/en/models/fable-5-1/overview, which
|
|
75
|
+
* moves claude-fable-5 to "Legacy models (still available)" → superseded.
|
|
76
|
+
* Same tier and same $10/$50 as fable-5, 1M ctx / 128K out, adaptive
|
|
77
|
+
* thinking always on, effort low|medium|high|xhigh|max default high, vision
|
|
78
|
+
* + tools + caching, reliable knowledge cutoff Jun 2026. The one price that
|
|
79
|
+
* is NOT the fleet ratio: cache read is $0.25/MTok — the pricing page
|
|
80
|
+
* footnotes fable-5-1 / mythos-5-1 as 0.025× input where every other model
|
|
81
|
+
* is 0.1×; cache write is the usual 1.25× ($12.50). Not modeled: the beta
|
|
82
|
+
* per-message effort / turn-scoped system messages / `display: "updates"`,
|
|
83
|
+
* and the three fable-5 → fable-5-1 breaking changes (forced tool_choice
|
|
84
|
+
* any/tool now 400s, thinking blocks are model-bound, editing earlier turns
|
|
85
|
+
* invalidates them) — Synthase sends tool_choice auto and is append-only,
|
|
86
|
+
* and the pre-commit dispatch probe covers the entry as sent.)
|
|
64
87
|
* - OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
|
|
65
88
|
* 2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
|
|
66
89
|
* 20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
|
|
@@ -97,12 +120,16 @@ import type { ModelDefinition } from './types.js';
|
|
|
97
120
|
* grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
98
121
|
* grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
99
122
|
* - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
100
|
-
* 2026-08-
|
|
123
|
+
* 2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
101
124
|
* never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
102
125
|
* has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
103
|
-
* the same card introduced
|
|
104
|
-
*
|
|
105
|
-
*
|
|
126
|
+
* the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
127
|
+
* unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
128
|
+
* 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
129
|
+
* where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
130
|
+
* `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
131
|
+
* also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
132
|
+
* deliberately not catalogued.)
|
|
106
133
|
* - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
107
134
|
* the US re-host (kimi-k3 flagship 2026-07-16
|
|
108
135
|
* — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
|
@@ -135,6 +162,16 @@ import type { ModelDefinition } from './types.js';
|
|
|
135
162
|
* $1.65/$4.951), so nothing would ever select it — and carrying both would
|
|
136
163
|
* put two selectable Alibaba flagships in one family. Revisit only if Alibaba
|
|
137
164
|
* publishes it as a distinct first-party DashScope model id.)
|
|
165
|
+
* (re-verified 2026-08-28 on the per-model pages, which publish USD rates per
|
|
166
|
+
* region directly — a better source than the CNY pricing table: qwen3.8-max's
|
|
167
|
+
* Singapore implicit-cache rate is $0.25, NOT the 20%/$0.40 the ZH cache doc's
|
|
168
|
+
* standard table implies. That doc names qwen3.8-max, qwen3.8-flash and
|
|
169
|
+
* qwen3.8-2.4t-a95b as exceptions and defers to the console; the model pages
|
|
170
|
+
* are the console's figures. qwen3.7-max's $0.50 (20%) is confirmed unchanged.
|
|
171
|
+
* qwen3.8-flash (2026-08-26) ADDED: $0.15/$0.47, implicit cache $0.016, 1M ctx
|
|
172
|
+
* / 131,072 out, Image+Text+Video input, tools, hybrid thinking ON by default
|
|
173
|
+
* with thinking_budget. It is cn-region-only — api.deepinfra.com has no
|
|
174
|
+
* Qwen/Qwen3.8-Flash — so ALIBABA_US_MODEL_MAP needs no entry.)
|
|
138
175
|
* - Zhipu: https://docs.z.ai/guides/overview/pricing + docs.z.ai/guides/llm/
|
|
139
176
|
* glm-5.3 (verified 2026-08-26: glm-5.3 shipped 2026-08-14 and IS in the
|
|
140
177
|
* catalog — $1.40/$4.40, cached $0.26, i.e. glm-5.2's card unchanged, 1M ctx,
|
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;
|
|
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"}
|
package/dist/models.js
CHANGED
|
@@ -7,6 +7,11 @@
|
|
|
7
7
|
*
|
|
8
8
|
* @module
|
|
9
9
|
*/
|
|
10
|
+
/**
|
|
11
|
+
* Monday–Friday as UTC weekday numbers, for `peakPricing` windows a provider
|
|
12
|
+
* prices by business day rather than by hour alone.
|
|
13
|
+
*/
|
|
14
|
+
const WEEKDAYS_UTC = [1, 2, 3, 4, 5];
|
|
10
15
|
/**
|
|
11
16
|
* All available AI models, grouped by provider, ordered from most to least capable.
|
|
12
17
|
*
|
|
@@ -60,6 +65,29 @@
|
|
|
60
65
|
* (re-verified 2026-08-06: added opus-4-7 — legacy but Active, $5/$25,
|
|
61
66
|
* cache $0.50/$6.25, 1M ctx / 128K out per the overview + pricing pages;
|
|
62
67
|
* models.dev first listed it 2026-08-06)
|
|
68
|
+
* (re-verified 2026-09-01 on /docs/en/about-claude/pricing: sonnet-5's
|
|
69
|
+
* $2/$10 launch pricing is now the STANDARD price — that page's own note
|
|
70
|
+
* reads "The previously scheduled increase to $3/$15 per million input/output
|
|
71
|
+
* tokens on September 1, 2026 will not occur." Catalog moved off the
|
|
72
|
+
* never-under-charge $3/$15 to $2/$10, cache $0.20/$2.50, and the
|
|
73
|
+
* KNOWN_DIVERGENCES entry was deleted. Every other Anthropic entry on that
|
|
74
|
+
* page is unchanged: fable-5 $10/$50 ($1/$12.50), opus-5 / opus-4-8 /
|
|
75
|
+
* opus-4-7 / opus-4-6 $5/$25 ($0.50/$6.25), sonnet-4-6 $3/$15 ($0.30/$3.75),
|
|
76
|
+
* haiku-4-5 $1/$5 ($0.10/$1.25).)
|
|
77
|
+
* (re-verified 2026-09-01 — ADDED claude-fable-5-1, released that day and
|
|
78
|
+
* listed as "Active (latest)" on /docs/en/models/fable-5-1/overview, which
|
|
79
|
+
* moves claude-fable-5 to "Legacy models (still available)" → superseded.
|
|
80
|
+
* Same tier and same $10/$50 as fable-5, 1M ctx / 128K out, adaptive
|
|
81
|
+
* thinking always on, effort low|medium|high|xhigh|max default high, vision
|
|
82
|
+
* + tools + caching, reliable knowledge cutoff Jun 2026. The one price that
|
|
83
|
+
* is NOT the fleet ratio: cache read is $0.25/MTok — the pricing page
|
|
84
|
+
* footnotes fable-5-1 / mythos-5-1 as 0.025× input where every other model
|
|
85
|
+
* is 0.1×; cache write is the usual 1.25× ($12.50). Not modeled: the beta
|
|
86
|
+
* per-message effort / turn-scoped system messages / `display: "updates"`,
|
|
87
|
+
* and the three fable-5 → fable-5-1 breaking changes (forced tool_choice
|
|
88
|
+
* any/tool now 400s, thinking blocks are model-bound, editing earlier turns
|
|
89
|
+
* invalidates them) — Synthase sends tool_choice auto and is append-only,
|
|
90
|
+
* and the pre-commit dispatch probe covers the entry as sent.)
|
|
63
91
|
* - OpenAI: https://developers.openai.com/api/docs/pricing (GPT-5.6 family GA
|
|
64
92
|
* 2026-07-09; REPRICED 2026-07-30: -luna cut 80% to $0.20/$1.20, -terra cut
|
|
65
93
|
* 20% to $2/$12, -sol unchanged $5/$30; cache read 0.1× input; gpt-5.5/
|
|
@@ -96,12 +124,16 @@
|
|
|
96
124
|
* grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
97
125
|
* grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
98
126
|
* - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
99
|
-
* 2026-08-
|
|
127
|
+
* 2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
100
128
|
* never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
101
129
|
* has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
102
|
-
* the same card introduced
|
|
103
|
-
*
|
|
104
|
-
*
|
|
130
|
+
* the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
131
|
+
* unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
132
|
+
* 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
133
|
+
* where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
134
|
+
* `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
135
|
+
* also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
136
|
+
* deliberately not catalogued.)
|
|
105
137
|
* - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
106
138
|
* the US re-host (kimi-k3 flagship 2026-07-16
|
|
107
139
|
* — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
|
@@ -134,6 +166,16 @@
|
|
|
134
166
|
* $1.65/$4.951), so nothing would ever select it — and carrying both would
|
|
135
167
|
* put two selectable Alibaba flagships in one family. Revisit only if Alibaba
|
|
136
168
|
* publishes it as a distinct first-party DashScope model id.)
|
|
169
|
+
* (re-verified 2026-08-28 on the per-model pages, which publish USD rates per
|
|
170
|
+
* region directly — a better source than the CNY pricing table: qwen3.8-max's
|
|
171
|
+
* Singapore implicit-cache rate is $0.25, NOT the 20%/$0.40 the ZH cache doc's
|
|
172
|
+
* standard table implies. That doc names qwen3.8-max, qwen3.8-flash and
|
|
173
|
+
* qwen3.8-2.4t-a95b as exceptions and defers to the console; the model pages
|
|
174
|
+
* are the console's figures. qwen3.7-max's $0.50 (20%) is confirmed unchanged.
|
|
175
|
+
* qwen3.8-flash (2026-08-26) ADDED: $0.15/$0.47, implicit cache $0.016, 1M ctx
|
|
176
|
+
* / 131,072 out, Image+Text+Video input, tools, hybrid thinking ON by default
|
|
177
|
+
* with thinking_budget. It is cn-region-only — api.deepinfra.com has no
|
|
178
|
+
* Qwen/Qwen3.8-Flash — so ALIBABA_US_MODEL_MAP needs no entry.)
|
|
137
179
|
* - Zhipu: https://docs.z.ai/guides/overview/pricing + docs.z.ai/guides/llm/
|
|
138
180
|
* glm-5.3 (verified 2026-08-26: glm-5.3 shipped 2026-08-14 and IS in the
|
|
139
181
|
* catalog — $1.40/$4.40, cached $0.26, i.e. glm-5.2's card unchanged, 1M ctx,
|
|
@@ -164,6 +206,41 @@ export const MODELS = [
|
|
|
164
206
|
// budget_tokens is REJECTED (400) on fable-5 / opus-4-8 / sonnet-5 and
|
|
165
207
|
// deprecated on the 4.6 family. Only Haiku 4.5 still uses budget_tokens.
|
|
166
208
|
// ---------------------------------------------------------------------------
|
|
209
|
+
{
|
|
210
|
+
id: 'claude-fable-5-1',
|
|
211
|
+
provider: 'anthropic',
|
|
212
|
+
label: 'Claude Fable 5.1',
|
|
213
|
+
description: 'Most capable Anthropic — frontier reasoning & long-horizon agents',
|
|
214
|
+
contextWindow: 1_000_000,
|
|
215
|
+
maxOutputTokens: 128_000,
|
|
216
|
+
supportsThinking: true,
|
|
217
|
+
thinkingBudgetTokens: 16_000,
|
|
218
|
+
thinkingConfigurable: true,
|
|
219
|
+
supportedEffortLevels: ['low', 'medium', 'high', 'xhigh', 'max'],
|
|
220
|
+
defaultEffortLevel: 'high',
|
|
221
|
+
// Thinking is ALWAYS ON (adaptive); effort is the only depth control, and
|
|
222
|
+
// all five levels are supported (effort docs, verified 2026-09-01).
|
|
223
|
+
// Default/recommended is high; xhigh/max for the most capability-sensitive
|
|
224
|
+
// agentic work, medium/low for routine work.
|
|
225
|
+
supportsVision: true,
|
|
226
|
+
supportsPromptCaching: true,
|
|
227
|
+
supportsTools: true,
|
|
228
|
+
webSearchToolType: 'web_search_20260209',
|
|
229
|
+
// Dynamic-filtering server tools, same as the rest of the 4.6+ Anthropic
|
|
230
|
+
// fleet. Not currently sent by Synthase beyond webSearchToolType
|
|
231
|
+
// (request-shape.ts forwards only that one).
|
|
232
|
+
codeExecutionToolType: 'code_execution_20260521',
|
|
233
|
+
webFetchToolType: 'web_fetch_20260209',
|
|
234
|
+
inputPricePerMTok: 10,
|
|
235
|
+
outputPricePerMTok: 50,
|
|
236
|
+
// Cache write is the standard 1.25× input. Cache READ is the documented
|
|
237
|
+
// exception to Anthropic's 0.1× rule: 0.025× input on fable-5-1 /
|
|
238
|
+
// mythos-5-1 only ("$0.25 / MTok", pricing page footnote 1).
|
|
239
|
+
cacheReadPricePerMTok: 0.25,
|
|
240
|
+
cacheWritePricePerMTok: 12.5,
|
|
241
|
+
// Reliable knowledge cutoff Jun 2026 (model page "Specifications").
|
|
242
|
+
knowledgeCutoff: '2026-06-01',
|
|
243
|
+
},
|
|
167
244
|
{
|
|
168
245
|
id: 'claude-fable-5',
|
|
169
246
|
provider: 'anthropic',
|
|
@@ -196,6 +273,11 @@ export const MODELS = [
|
|
|
196
273
|
cacheReadPricePerMTok: 1,
|
|
197
274
|
cacheWritePricePerMTok: 12.5,
|
|
198
275
|
knowledgeCutoff: '2026-01-01',
|
|
276
|
+
// Superseded by claude-fable-5-1 (same tier, same $10/$50 input/output,
|
|
277
|
+
// cheaper cache reads); Anthropic now lists it under "Legacy models (still
|
|
278
|
+
// available)". Still served and priceable, just not OFFERED.
|
|
279
|
+
deprecatedAt: '2026-09-01',
|
|
280
|
+
supersededBy: 'claude-fable-5-1',
|
|
199
281
|
},
|
|
200
282
|
{
|
|
201
283
|
id: 'claude-opus-5',
|
|
@@ -304,13 +386,14 @@ export const MODELS = [
|
|
|
304
386
|
// Synthase (request-shape.ts forwards only webSearchToolType).
|
|
305
387
|
codeExecutionToolType: 'code_execution_20260521',
|
|
306
388
|
webFetchToolType: 'web_fetch_20260209',
|
|
307
|
-
//
|
|
308
|
-
//
|
|
309
|
-
|
|
310
|
-
|
|
389
|
+
// The launch "intro" pricing IS the standard price as of 2026-09-01 — the
|
|
390
|
+
// pricing page's own note says the scheduled 2026-09-01 rise to $3/$15
|
|
391
|
+
// will not occur (verified 2026-09-01).
|
|
392
|
+
inputPricePerMTok: 2,
|
|
393
|
+
outputPricePerMTok: 10,
|
|
311
394
|
// Anthropic 5-minute prompt cache: read 0.1× input, write 1.25× input.
|
|
312
|
-
cacheReadPricePerMTok: 0.
|
|
313
|
-
cacheWritePricePerMTok:
|
|
395
|
+
cacheReadPricePerMTok: 0.2,
|
|
396
|
+
cacheWritePricePerMTok: 2.5,
|
|
314
397
|
knowledgeCutoff: '2026-01-01',
|
|
315
398
|
},
|
|
316
399
|
{
|
|
@@ -716,7 +799,7 @@ export const MODELS = [
|
|
|
716
799
|
// "New Stable" 2026-08-13. LIST price $1.50/$7.50 — same as 3.6-flash.
|
|
717
800
|
// Google runs a launch promo ($0.75/$3.75, cache read $0.075) through
|
|
718
801
|
// 2026-12-31; billed here at standard list so metering never under-charges
|
|
719
|
-
// (same policy as
|
|
802
|
+
// (same policy as gpt-5.6-sol's promo card — see the matching
|
|
720
803
|
// KNOWN_DIVERGENCES entry in scripts/check-model-freshness.mjs, expiring
|
|
721
804
|
// 2026-12-31).
|
|
722
805
|
inputPricePerMTok: 1.5,
|
|
@@ -1019,12 +1102,16 @@ export const MODELS = [
|
|
|
1019
1102
|
// input cost rose far more than the list prices suggest. Both are free-tier
|
|
1020
1103
|
// models (flash is `freeTier`, pro is the free-tier planner).
|
|
1021
1104
|
// The two post-landing re-verifications, both settled 2026-08-18:
|
|
1022
|
-
// 1. WEEKDAYS OR DAILY — DAILY
|
|
1023
|
-
// are 01:00 - 04:00 and 06:00 - 10:00
|
|
1024
|
-
//
|
|
1025
|
-
//
|
|
1026
|
-
//
|
|
1027
|
-
//
|
|
1105
|
+
// 1. WEEKDAYS OR DAILY — DAILY on 2026-08-18, WEEKDAYS as of 2026-08-31.
|
|
1106
|
+
// The card now reads "Peak hours are 01:00 - 04:00 and 06:00 - 10:00
|
|
1107
|
+
// UTC, Monday through Friday (all other hours are off-peak)"; on
|
|
1108
|
+
// 2026-08-18 the same sentence carried no day qualifier (press coverage
|
|
1109
|
+
// had described them as weekday-only, and it turned out to be right, or
|
|
1110
|
+
// to have become right). That is the case the 2026-08-18 note said would
|
|
1111
|
+
// need the FIELD extended rather than the numbers nudged — so
|
|
1112
|
+
// `peakPricing` windows gained `daysOfWeekUtc` and both entries carry
|
|
1113
|
+
// WEEKDAYS_UTC. Modeling them as daily bills a 2× on ~29 weekend hours a
|
|
1114
|
+
// week that DeepSeek does not charge.
|
|
1028
1115
|
// 2. THE CN-VS-US DEFAULT — settled per model, and they differ. Post-rise
|
|
1029
1116
|
// DeepInfra's US flash rates (0.08/0.18/0.016) are BELOW CN's off-peak
|
|
1030
1117
|
// on both input and output, so Flash defaults US; Pro's US re-host still
|
|
@@ -1090,14 +1177,15 @@ export const MODELS = [
|
|
|
1090
1177
|
},
|
|
1091
1178
|
// The peak-hour 2× surcharge is ON the rate card and LIVE since
|
|
1092
1179
|
// 2026-08-16T16:00Z. Peak = 01:00-04:00 and 06:00-10:00 UTC (Beijing
|
|
1093
|
-
// business hours)
|
|
1094
|
-
// like these were once pre-wired ahead of the card and
|
|
1095
|
-
// peak-window turn for weeks — hence the rule that they
|
|
1096
|
-
// once the provider's own page shows them, which it now
|
|
1180
|
+
// business hours) MONDAY THROUGH FRIDAY, at exactly 2× the off-peak rates
|
|
1181
|
+
// above. Windows like these were once pre-wired ahead of the card and
|
|
1182
|
+
// over-billed every peak-window turn for weeks — hence the rule that they
|
|
1183
|
+
// only exist here once the provider's own page shows them, which it now
|
|
1184
|
+
// does.
|
|
1097
1185
|
peakPricing: {
|
|
1098
1186
|
windows: [
|
|
1099
|
-
{ startMinuteUtc: 60, endMinuteUtc: 240 },
|
|
1100
|
-
{ startMinuteUtc: 360, endMinuteUtc: 600 },
|
|
1187
|
+
{ startMinuteUtc: 60, endMinuteUtc: 240, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1188
|
+
{ startMinuteUtc: 360, endMinuteUtc: 600, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1101
1189
|
],
|
|
1102
1190
|
multiplier: 2,
|
|
1103
1191
|
},
|
|
@@ -1151,13 +1239,13 @@ export const MODELS = [
|
|
|
1151
1239
|
regionPricing: {
|
|
1152
1240
|
us: { inputPricePerMTok: 0.08, outputPricePerMTok: 0.18, cacheReadPricePerMTok: 0.016 },
|
|
1153
1241
|
},
|
|
1154
|
-
// Peak-hour surcharge live since 2026-08-16T16:00Z (see
|
|
1155
|
-
// It applies to the NATIVE CN card only — this model
|
|
1156
|
-
// re-host, which is flat, so most turns never take it.
|
|
1242
|
+
// Peak-hour surcharge live since 2026-08-16T16:00Z, Mon-Fri (see
|
|
1243
|
+
// deepseek-v4-pro). It applies to the NATIVE CN card only — this model
|
|
1244
|
+
// defaults to the US re-host, which is flat, so most turns never take it.
|
|
1157
1245
|
peakPricing: {
|
|
1158
1246
|
windows: [
|
|
1159
|
-
{ startMinuteUtc: 60, endMinuteUtc: 240 },
|
|
1160
|
-
{ startMinuteUtc: 360, endMinuteUtc: 600 },
|
|
1247
|
+
{ startMinuteUtc: 60, endMinuteUtc: 240, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1248
|
+
{ startMinuteUtc: 360, endMinuteUtc: 600, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1161
1249
|
],
|
|
1162
1250
|
multiplier: 2,
|
|
1163
1251
|
},
|
|
@@ -1469,16 +1557,24 @@ export const MODELS = [
|
|
|
1469
1557
|
supportedEffortLevels: ['4K', '8K', '16K', '32K'],
|
|
1470
1558
|
defaultEffortLevel: '8K',
|
|
1471
1559
|
effortBudgetTokens: { '4K': 4000, '8K': 8000, '16K': 16000, '32K': 32000 },
|
|
1472
|
-
//
|
|
1473
|
-
//
|
|
1474
|
-
// false
|
|
1560
|
+
// The provider's own model page DOES list Image/Text/Video input (checked
|
|
1561
|
+
// 2026-08-28, superseding the earlier "text generation only" reading of the
|
|
1562
|
+
// model catalog page). Kept false anyway: this model's DEFAULT region is the
|
|
1563
|
+
// DeepInfra US re-host, which serves it as `text-generation` — vision is not
|
|
1564
|
+
// offerable where it actually dispatches. Native-only qwen3.8-flash below
|
|
1565
|
+
// carries the flag.
|
|
1475
1566
|
supportsVision: false,
|
|
1476
1567
|
supportsPromptCaching: true,
|
|
1477
1568
|
supportsTools: true,
|
|
1478
1569
|
inputPricePerMTok: 2,
|
|
1479
1570
|
outputPricePerMTok: 6,
|
|
1480
|
-
// Implicit context cache
|
|
1481
|
-
|
|
1571
|
+
// Implicit context cache. The ZH cache doc's standard 20% does NOT apply
|
|
1572
|
+
// here: that doc names qwen3.8-max as an exception and points at the
|
|
1573
|
+
// console, and the model page publishes the real Singapore rate — $0.25,
|
|
1574
|
+
// i.e. 12.5% of input (verified 2026-08-28 on
|
|
1575
|
+
// alibabacloud.com/help/en/model-studio/qwen3-8-max). Implicit creation is
|
|
1576
|
+
// not billed beyond input, so write stays 1× input.
|
|
1577
|
+
cacheReadPricePerMTok: 0.25,
|
|
1482
1578
|
cacheWritePricePerMTok: 2,
|
|
1483
1579
|
regions: ['us', 'cn'],
|
|
1484
1580
|
// US = DeepInfra (Qwen/Qwen3.8-Max), verified 2026-08-14 against
|
|
@@ -1492,6 +1588,49 @@ export const MODELS = [
|
|
|
1492
1588
|
// Not published by Alibaba — best-effort estimate.
|
|
1493
1589
|
knowledgeCutoff: '2026-04-01',
|
|
1494
1590
|
},
|
|
1591
|
+
{
|
|
1592
|
+
id: 'qwen3.8-flash',
|
|
1593
|
+
provider: 'alibaba',
|
|
1594
|
+
label: 'Qwen3.8 Flash',
|
|
1595
|
+
description: 'Alibaba cheap tier — 1M context, multimodal, hybrid thinking',
|
|
1596
|
+
// Model page: context window 1,000,000; max input 991,808 (983,616 in
|
|
1597
|
+
// thinking mode); max chain-of-thought 262,144; max output 131,072.
|
|
1598
|
+
contextWindow: 1_000_000,
|
|
1599
|
+
maxOutputTokens: 131_072,
|
|
1600
|
+
// Same hybrid-thinking mechanism as qwen3.8-max — the deep-thinking doc
|
|
1601
|
+
// lists the "Qwen3.8 Flash series" as hybrid with thinking ON by default,
|
|
1602
|
+
// and thinking_budget applies — so effort scales the budget identically.
|
|
1603
|
+
supportsThinking: true,
|
|
1604
|
+
thinkingBudgetTokens: 8_000,
|
|
1605
|
+
thinkingConfigurable: true,
|
|
1606
|
+
supportedEffortLevels: ['4K', '8K', '16K', '32K'],
|
|
1607
|
+
defaultEffortLevel: '8K',
|
|
1608
|
+
effortBudgetTokens: { '4K': 4000, '8K': 8000, '16K': 16000, '32K': 32000 },
|
|
1609
|
+
// Model page lists Image / Text / Video input. Unlike qwen3.8-max this one
|
|
1610
|
+
// has no US re-host to lose it to — it only ever dispatches to the native
|
|
1611
|
+
// host, which serves the multimodal surface.
|
|
1612
|
+
supportsVision: true,
|
|
1613
|
+
supportsPromptCaching: true,
|
|
1614
|
+
supportsTools: true,
|
|
1615
|
+
// Singapore International list card, published in USD on the model page:
|
|
1616
|
+
// $0.15 / $0.47, implicit cache $0.016. (Beijing is cheaper at
|
|
1617
|
+
// $0.113/$0.382/$0.014; the bond calls the international endpoint.)
|
|
1618
|
+
inputPricePerMTok: 0.15,
|
|
1619
|
+
outputPricePerMTok: 0.47,
|
|
1620
|
+
// Implicit context cache. Like qwen3.8-max this model is an explicit
|
|
1621
|
+
// exception to the ZH doc's standard 20%; the published rate is $0.016
|
|
1622
|
+
// (~10.7% of input). Creation is not billed beyond input → write = 1×.
|
|
1623
|
+
cacheReadPricePerMTok: 0.016,
|
|
1624
|
+
cacheWritePricePerMTok: 0.15,
|
|
1625
|
+
// No US re-host exists — DeepInfra serves Qwen3.8-Max, Qwen3.8-27B and
|
|
1626
|
+
// Qwen3.8-2.4T-A95B but returns "model not found" for Qwen/Qwen3.8-Flash
|
|
1627
|
+
// (checked 2026-08-28), so this is pinned to the native host and bills the
|
|
1628
|
+
// card above. ALIBABA_US_MODEL_MAP therefore needs no entry.
|
|
1629
|
+
regions: ['cn'],
|
|
1630
|
+
// Not published by Alibaba — best-effort estimate, same 3.8 generation as
|
|
1631
|
+
// qwen3.8-max.
|
|
1632
|
+
knowledgeCutoff: '2026-04-01',
|
|
1633
|
+
},
|
|
1495
1634
|
{
|
|
1496
1635
|
id: 'qwen3.7-max',
|
|
1497
1636
|
provider: 'alibaba',
|
package/dist/types.d.ts
CHANGED
|
@@ -224,11 +224,22 @@ export interface ModelDefinition {
|
|
|
224
224
|
*
|
|
225
225
|
* Windows are minutes-since-midnight UTC, half-open `[start, end)`; a window
|
|
226
226
|
* may wrap midnight (`start > end`).
|
|
227
|
+
*
|
|
228
|
+
* `daysOfWeekUtc` restricts a window to certain UTC weekdays (`0` = Sunday …
|
|
229
|
+
* `6` = Saturday); omitted means every day. A provider that prices its peak
|
|
230
|
+
* by BUSINESS hours qualifies them by day, and a day-less window over-bills
|
|
231
|
+
* every weekend hour inside it at the full multiplier — DeepSeek's card
|
|
232
|
+
* carried no day qualifier when these windows landed on 2026-08-16 and added
|
|
233
|
+
* "Monday through Friday" by 2026-08-31, which is exactly the 2× weekend
|
|
234
|
+
* over-bill this field exists to prevent. A wrapping window belongs to the
|
|
235
|
+
* day it STARTS on, so its post-midnight tail is still matched against the
|
|
236
|
+
* previous day.
|
|
227
237
|
*/
|
|
228
238
|
peakPricing?: {
|
|
229
239
|
windows: {
|
|
230
240
|
startMinuteUtc: number;
|
|
231
241
|
endMinuteUtc: number;
|
|
242
|
+
daysOfWeekUtc?: number[];
|
|
232
243
|
}[];
|
|
233
244
|
multiplier: number;
|
|
234
245
|
};
|
|
@@ -280,6 +291,7 @@ export interface ModelDefinition {
|
|
|
280
291
|
windows: {
|
|
281
292
|
startMinuteUtc: number;
|
|
282
293
|
endMinuteUtc: number;
|
|
294
|
+
daysOfWeekUtc?: number[];
|
|
283
295
|
}[];
|
|
284
296
|
multiplier: number;
|
|
285
297
|
};
|
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;;;;;;;;;;;;;GAaG;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;;;;;;;;;;;;;;;;;;;OAmBG;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;;;;;;;;;;;;;;;;;;;OAmBG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAA;IAClC;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B;;;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
|
|
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;;;;;;;;;;;;;GAaG;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;;;;;;;;;;;;;;;;;;;OAmBG;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;;;;;;;;;;;;;;;;;;;OAmBG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAA;IAClC;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B;;;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;;;;;;;;;;;;;;;;;;;;OAoBG;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;KACnB,CAAA;IACD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OA+BG;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;SACnB,CAAA;QACD,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,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