@molecule/api-resource-ai-models 1.3.0 → 1.5.0
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 +61 -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 +40 -4
- package/dist/models.d.ts.map +1 -1
- package/dist/models.js +171 -36
- 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: 2026-09-02T19:41:05.247Z
|
|
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/
|
|
@@ -807,18 +846,31 @@ Sources (verified 2026-07-28; OpenAI re-verified 2026-07-31 after the
|
|
|
807
846
|
at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
|
|
808
847
|
never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
|
|
809
848
|
2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
|
|
849
|
+
(re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
|
|
850
|
+
3.7-flash, which the models page now calls "previous-generation". THIRD id
|
|
851
|
+
on the same flash-tier promo row, verbatim "$0.75 through December 31,
|
|
852
|
+
2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
|
|
853
|
+
$0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
|
|
854
|
+
Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
|
|
855
|
+
low|medium|high, vision, tools, caching, search grounding, code execution,
|
|
856
|
+
url context — identical surface to 3.7-flash. Knowledge cutoff is published
|
|
857
|
+
by neither Google nor models.dev for this id)
|
|
810
858
|
- xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
|
|
811
859
|
(grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
|
|
812
860
|
not modeled; reasoning_effort low|medium|high default high, image input;
|
|
813
861
|
grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
814
862
|
grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
815
863
|
- DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
816
|
-
2026-08-
|
|
864
|
+
2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
817
865
|
never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
818
866
|
has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
819
|
-
the same card introduced
|
|
820
|
-
|
|
821
|
-
|
|
867
|
+
the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
868
|
+
unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
869
|
+
06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
870
|
+
where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
871
|
+
`daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
872
|
+
also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
873
|
+
deliberately not catalogued.)
|
|
822
874
|
- Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
823
875
|
the US re-host (kimi-k3 flagship 2026-07-16
|
|
824
876
|
— 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
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/
|
|
@@ -91,18 +114,31 @@ import type { ModelDefinition } from './types.js';
|
|
|
91
114
|
* at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
|
|
92
115
|
* never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
|
|
93
116
|
* 2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
|
|
117
|
+
* (re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
|
|
118
|
+
* 3.7-flash, which the models page now calls "previous-generation". THIRD id
|
|
119
|
+
* on the same flash-tier promo row, verbatim "$0.75 through December 31,
|
|
120
|
+
* 2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
|
|
121
|
+
* $0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
|
|
122
|
+
* Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
|
|
123
|
+
* low|medium|high, vision, tools, caching, search grounding, code execution,
|
|
124
|
+
* url context — identical surface to 3.7-flash. Knowledge cutoff is published
|
|
125
|
+
* by neither Google nor models.dev for this id)
|
|
94
126
|
* - xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
|
|
95
127
|
* (grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
|
|
96
128
|
* not modeled; reasoning_effort low|medium|high default high, image input;
|
|
97
129
|
* grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
98
130
|
* grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
99
131
|
* - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
100
|
-
* 2026-08-
|
|
132
|
+
* 2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
101
133
|
* never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
102
134
|
* has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
103
|
-
* the same card introduced
|
|
104
|
-
*
|
|
105
|
-
*
|
|
135
|
+
* the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
136
|
+
* unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
137
|
+
* 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
138
|
+
* where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
139
|
+
* `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
140
|
+
* also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
141
|
+
* deliberately not catalogued.)
|
|
106
142
|
* - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
107
143
|
* the US re-host (kimi-k3 flagship 2026-07-16
|
|
108
144
|
* — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
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;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiMG;AACH,eAAO,MAAM,MAAM,EAAE,SAAS,eAAe,EAusDnC,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/
|
|
@@ -90,18 +118,31 @@
|
|
|
90
118
|
* at the post-promo list rate ($1.50/$7.50, cache read $0.15) under the same
|
|
91
119
|
* never-under-charge policy, each with a KNOWN_DIVERGENCES entry expiring
|
|
92
120
|
* 2026-12-31. gemini-3.5-flash carries no promo: still $1.50/$9.00, $0.15)
|
|
121
|
+
* (re-verified 2026-09-02: gemini-3.8-flash "New Stable" — supersedes
|
|
122
|
+
* 3.7-flash, which the models page now calls "previous-generation". THIRD id
|
|
123
|
+
* on the same flash-tier promo row, verbatim "$0.75 through December 31,
|
|
124
|
+
* 2026. $1.50 starting January 1, 2027." / "$3.75 … $7.50" / "$0.075 …
|
|
125
|
+
* $0.15", so it too is cataloged at list with a KNOWN_DIVERGENCES entry.
|
|
126
|
+
* Specs from /docs/models/gemini-3.8-flash: 1M ctx / 65,536 out, thinking
|
|
127
|
+
* low|medium|high, vision, tools, caching, search grounding, code execution,
|
|
128
|
+
* url context — identical surface to 3.7-flash. Knowledge cutoff is published
|
|
129
|
+
* by neither Google nor models.dev for this id)
|
|
93
130
|
* - xAI: https://docs.x.ai/developers/models + /developers/grok-4-5
|
|
94
131
|
* (grok-4.5 flagship 2026-07-08: $2/$6, 500K ctx, ≥200K prompts bill 2× —
|
|
95
132
|
* not modeled; reasoning_effort low|medium|high default high, image input;
|
|
96
133
|
* grok-4.3 still served at $1.25/$2.50 with the bigger 1M window;
|
|
97
134
|
* grok-code-fast-1 no longer listed — retires 2026-08-15)
|
|
98
135
|
* - DeepSeek: https://api-docs.deepseek.com/quick_start/pricing (verified
|
|
99
|
-
* 2026-08-
|
|
136
|
+
* 2026-08-31; legacy deepseek-chat/-reasoner ids fully retired 2026-07-24 —
|
|
100
137
|
* never in this catalog. The V4-Pro-GA price RISE effective 2026-08-16T16:00Z
|
|
101
138
|
* has LANDED and is folded into the base fields, along with the peak-hour 2×
|
|
102
|
-
* the same card introduced
|
|
103
|
-
*
|
|
104
|
-
*
|
|
139
|
+
* the same card introduced; every rate re-read on the card 2026-08-31 and
|
|
140
|
+
* unchanged. The card now qualifies the windows BY DAY — "01:00 - 04:00 and
|
|
141
|
+
* 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)" —
|
|
142
|
+
* where on 2026-08-18 it carried no day qualifier at all, so the windows are
|
|
143
|
+
* `daysOfWeekUtc`-restricted rather than daily. `deepseek-v4-flash-vision-exp`
|
|
144
|
+
* also appears on the card at flash's rates: EXPERIMENTAL and vision-only-new,
|
|
145
|
+
* deliberately not catalogued.)
|
|
105
146
|
* - Moonshot: https://platform.kimi.ai/docs/models + DeepInfra's model API for
|
|
106
147
|
* the US re-host (kimi-k3 flagship 2026-07-16
|
|
107
148
|
* — 2.8T MoE, 1M ctx, $3/$15 — NOT added: thinking is forced-on with
|
|
@@ -174,6 +215,41 @@ export const MODELS = [
|
|
|
174
215
|
// budget_tokens is REJECTED (400) on fable-5 / opus-4-8 / sonnet-5 and
|
|
175
216
|
// deprecated on the 4.6 family. Only Haiku 4.5 still uses budget_tokens.
|
|
176
217
|
// ---------------------------------------------------------------------------
|
|
218
|
+
{
|
|
219
|
+
id: 'claude-fable-5-1',
|
|
220
|
+
provider: 'anthropic',
|
|
221
|
+
label: 'Claude Fable 5.1',
|
|
222
|
+
description: 'Most capable Anthropic — frontier reasoning & long-horizon agents',
|
|
223
|
+
contextWindow: 1_000_000,
|
|
224
|
+
maxOutputTokens: 128_000,
|
|
225
|
+
supportsThinking: true,
|
|
226
|
+
thinkingBudgetTokens: 16_000,
|
|
227
|
+
thinkingConfigurable: true,
|
|
228
|
+
supportedEffortLevels: ['low', 'medium', 'high', 'xhigh', 'max'],
|
|
229
|
+
defaultEffortLevel: 'high',
|
|
230
|
+
// Thinking is ALWAYS ON (adaptive); effort is the only depth control, and
|
|
231
|
+
// all five levels are supported (effort docs, verified 2026-09-01).
|
|
232
|
+
// Default/recommended is high; xhigh/max for the most capability-sensitive
|
|
233
|
+
// agentic work, medium/low for routine work.
|
|
234
|
+
supportsVision: true,
|
|
235
|
+
supportsPromptCaching: true,
|
|
236
|
+
supportsTools: true,
|
|
237
|
+
webSearchToolType: 'web_search_20260209',
|
|
238
|
+
// Dynamic-filtering server tools, same as the rest of the 4.6+ Anthropic
|
|
239
|
+
// fleet. Not currently sent by Synthase beyond webSearchToolType
|
|
240
|
+
// (request-shape.ts forwards only that one).
|
|
241
|
+
codeExecutionToolType: 'code_execution_20260521',
|
|
242
|
+
webFetchToolType: 'web_fetch_20260209',
|
|
243
|
+
inputPricePerMTok: 10,
|
|
244
|
+
outputPricePerMTok: 50,
|
|
245
|
+
// Cache write is the standard 1.25× input. Cache READ is the documented
|
|
246
|
+
// exception to Anthropic's 0.1× rule: 0.025× input on fable-5-1 /
|
|
247
|
+
// mythos-5-1 only ("$0.25 / MTok", pricing page footnote 1).
|
|
248
|
+
cacheReadPricePerMTok: 0.25,
|
|
249
|
+
cacheWritePricePerMTok: 12.5,
|
|
250
|
+
// Reliable knowledge cutoff Jun 2026 (model page "Specifications").
|
|
251
|
+
knowledgeCutoff: '2026-06-01',
|
|
252
|
+
},
|
|
177
253
|
{
|
|
178
254
|
id: 'claude-fable-5',
|
|
179
255
|
provider: 'anthropic',
|
|
@@ -206,6 +282,11 @@ export const MODELS = [
|
|
|
206
282
|
cacheReadPricePerMTok: 1,
|
|
207
283
|
cacheWritePricePerMTok: 12.5,
|
|
208
284
|
knowledgeCutoff: '2026-01-01',
|
|
285
|
+
// Superseded by claude-fable-5-1 (same tier, same $10/$50 input/output,
|
|
286
|
+
// cheaper cache reads); Anthropic now lists it under "Legacy models (still
|
|
287
|
+
// available)". Still served and priceable, just not OFFERED.
|
|
288
|
+
deprecatedAt: '2026-09-01',
|
|
289
|
+
supersededBy: 'claude-fable-5-1',
|
|
209
290
|
},
|
|
210
291
|
{
|
|
211
292
|
id: 'claude-opus-5',
|
|
@@ -314,13 +395,14 @@ export const MODELS = [
|
|
|
314
395
|
// Synthase (request-shape.ts forwards only webSearchToolType).
|
|
315
396
|
codeExecutionToolType: 'code_execution_20260521',
|
|
316
397
|
webFetchToolType: 'web_fetch_20260209',
|
|
317
|
-
//
|
|
318
|
-
//
|
|
319
|
-
|
|
320
|
-
|
|
398
|
+
// The launch "intro" pricing IS the standard price as of 2026-09-01 — the
|
|
399
|
+
// pricing page's own note says the scheduled 2026-09-01 rise to $3/$15
|
|
400
|
+
// will not occur (verified 2026-09-01).
|
|
401
|
+
inputPricePerMTok: 2,
|
|
402
|
+
outputPricePerMTok: 10,
|
|
321
403
|
// Anthropic 5-minute prompt cache: read 0.1× input, write 1.25× input.
|
|
322
|
-
cacheReadPricePerMTok: 0.
|
|
323
|
-
cacheWritePricePerMTok:
|
|
404
|
+
cacheReadPricePerMTok: 0.2,
|
|
405
|
+
cacheWritePricePerMTok: 2.5,
|
|
324
406
|
knowledgeCutoff: '2026-01-01',
|
|
325
407
|
},
|
|
326
408
|
{
|
|
@@ -703,11 +785,48 @@ export const MODELS = [
|
|
|
703
785
|
// replace outright: the google bond has never been implemented/wired, so no
|
|
704
786
|
// historical usage can reference the old ids.
|
|
705
787
|
// ---------------------------------------------------------------------------
|
|
788
|
+
{
|
|
789
|
+
id: 'gemini-3.8-flash',
|
|
790
|
+
provider: 'google',
|
|
791
|
+
label: 'Gemini 3.8 Flash',
|
|
792
|
+
description: 'Google agentic flagship — long-horizon engineering & autonomous agents',
|
|
793
|
+
// Verified against /docs/models/gemini-3.8-flash (2026-09-02) — same limits
|
|
794
|
+
// and capability surface as 3.7-flash.
|
|
795
|
+
contextWindow: 1_048_576,
|
|
796
|
+
maxOutputTokens: 65_536,
|
|
797
|
+
supportsThinking: true,
|
|
798
|
+
thinkingBudgetTokens: 10_000,
|
|
799
|
+
thinkingConfigurable: true,
|
|
800
|
+
// thinking_level low|medium|high — minimal NOT supported on this model.
|
|
801
|
+
supportedEffortLevels: ['low', 'medium', 'high'],
|
|
802
|
+
defaultEffortLevel: 'medium',
|
|
803
|
+
supportsVision: true,
|
|
804
|
+
supportsPromptCaching: true,
|
|
805
|
+
supportsTools: true,
|
|
806
|
+
webSearchToolType: 'google_search',
|
|
807
|
+
codeExecutionToolType: 'code_execution',
|
|
808
|
+
webFetchToolType: 'url_context',
|
|
809
|
+
// "New Stable" 2026-09-02. LIST price $1.50/$7.50 — same card as 3.7- and
|
|
810
|
+
// 3.6-flash, and on the same flash-tier launch promo ($0.75/$3.75, cache
|
|
811
|
+
// read $0.075) through 2026-12-31; billed here at standard list so metering
|
|
812
|
+
// never under-charges (identical policy and expiry to the two entries
|
|
813
|
+
// below — see the matching KNOWN_DIVERGENCES entry in
|
|
814
|
+
// scripts/check-model-freshness.mjs).
|
|
815
|
+
inputPricePerMTok: 1.5,
|
|
816
|
+
outputPricePerMTok: 7.5,
|
|
817
|
+
// Gemini context cache: read $0.15/M (0.1× input), no write premium
|
|
818
|
+
// (storage billed separately per hour — not modeled).
|
|
819
|
+
cacheReadPricePerMTok: 0.15,
|
|
820
|
+
cacheWritePricePerMTok: 1.5,
|
|
821
|
+
// Published by neither Google's model page nor models.dev for this id —
|
|
822
|
+
// carried from 3.7-flash (same generation) as a best-effort estimate.
|
|
823
|
+
knowledgeCutoff: '2026-03-01',
|
|
824
|
+
},
|
|
706
825
|
{
|
|
707
826
|
id: 'gemini-3.7-flash',
|
|
708
827
|
provider: 'google',
|
|
709
828
|
label: 'Gemini 3.7 Flash',
|
|
710
|
-
description: 'Google agentic flagship — complex coding & multi-step execution',
|
|
829
|
+
description: 'Previous Google agentic flagship — complex coding & multi-step execution',
|
|
711
830
|
// Verified against /docs/models/gemini-3.7-flash (2026-08-13).
|
|
712
831
|
contextWindow: 1_048_576,
|
|
713
832
|
maxOutputTokens: 65_536,
|
|
@@ -726,7 +845,7 @@ export const MODELS = [
|
|
|
726
845
|
// "New Stable" 2026-08-13. LIST price $1.50/$7.50 — same as 3.6-flash.
|
|
727
846
|
// Google runs a launch promo ($0.75/$3.75, cache read $0.075) through
|
|
728
847
|
// 2026-12-31; billed here at standard list so metering never under-charges
|
|
729
|
-
// (same policy as
|
|
848
|
+
// (same policy as gpt-5.6-sol's promo card — see the matching
|
|
730
849
|
// KNOWN_DIVERGENCES entry in scripts/check-model-freshness.mjs, expiring
|
|
731
850
|
// 2026-12-31).
|
|
732
851
|
inputPricePerMTok: 1.5,
|
|
@@ -737,6 +856,11 @@ export const MODELS = [
|
|
|
737
856
|
cacheWritePricePerMTok: 1.5,
|
|
738
857
|
// Not on Google's docs — models.dev reports 2026-03 (lead, not authority).
|
|
739
858
|
knowledgeCutoff: '2026-03-01',
|
|
859
|
+
// Superseded by gemini-3.8-flash (2026-09-02) — same flash tier, same list
|
|
860
|
+
// price; Google's own models page now calls 3.7 "previous-generation".
|
|
861
|
+
// Still served upstream, so it stays priceable.
|
|
862
|
+
deprecatedAt: '2026-09-02',
|
|
863
|
+
supersededBy: 'gemini-3.8-flash',
|
|
740
864
|
},
|
|
741
865
|
{
|
|
742
866
|
id: 'gemini-3.6-flash',
|
|
@@ -777,9 +901,11 @@ export const MODELS = [
|
|
|
777
901
|
knowledgeCutoff: '2026-01-01',
|
|
778
902
|
// Superseded by gemini-3.7-flash (2026-08-13) — same flash tier, same list
|
|
779
903
|
// price; Google's own models page now calls 3.6 "previous-generation".
|
|
780
|
-
// Still served upstream, so it stays priceable.
|
|
904
|
+
// Still served upstream, so it stays priceable. Re-pointed to
|
|
905
|
+
// gemini-3.8-flash on 2026-09-02 when 3.7 was itself deprecated —
|
|
906
|
+
// supersededBy must target a SELECTABLE model, never a chain.
|
|
781
907
|
deprecatedAt: '2026-08-13',
|
|
782
|
-
supersededBy: 'gemini-3.
|
|
908
|
+
supersededBy: 'gemini-3.8-flash',
|
|
783
909
|
},
|
|
784
910
|
{
|
|
785
911
|
id: 'gemini-3.5-flash',
|
|
@@ -810,11 +936,11 @@ export const MODELS = [
|
|
|
810
936
|
cacheWritePricePerMTok: 1.5,
|
|
811
937
|
knowledgeCutoff: '2025-01-01',
|
|
812
938
|
// Superseded within the flash tier (first by 3.6-flash on 2026-07-21, now
|
|
813
|
-
// pointed one hop to gemini-3.
|
|
939
|
+
// pointed one hop to gemini-3.8-flash — supersededBy must target a
|
|
814
940
|
// SELECTABLE model, never a chain). Still served upstream, so it stays
|
|
815
941
|
// priceable.
|
|
816
942
|
deprecatedAt: '2026-07-21',
|
|
817
|
-
supersededBy: 'gemini-3.
|
|
943
|
+
supersededBy: 'gemini-3.8-flash',
|
|
818
944
|
},
|
|
819
945
|
{
|
|
820
946
|
id: 'gemini-3.1-pro-preview',
|
|
@@ -1029,12 +1155,16 @@ export const MODELS = [
|
|
|
1029
1155
|
// input cost rose far more than the list prices suggest. Both are free-tier
|
|
1030
1156
|
// models (flash is `freeTier`, pro is the free-tier planner).
|
|
1031
1157
|
// The two post-landing re-verifications, both settled 2026-08-18:
|
|
1032
|
-
// 1. WEEKDAYS OR DAILY — DAILY
|
|
1033
|
-
// are 01:00 - 04:00 and 06:00 - 10:00
|
|
1034
|
-
//
|
|
1035
|
-
//
|
|
1036
|
-
//
|
|
1037
|
-
//
|
|
1158
|
+
// 1. WEEKDAYS OR DAILY — DAILY on 2026-08-18, WEEKDAYS as of 2026-08-31.
|
|
1159
|
+
// The card now reads "Peak hours are 01:00 - 04:00 and 06:00 - 10:00
|
|
1160
|
+
// UTC, Monday through Friday (all other hours are off-peak)"; on
|
|
1161
|
+
// 2026-08-18 the same sentence carried no day qualifier (press coverage
|
|
1162
|
+
// had described them as weekday-only, and it turned out to be right, or
|
|
1163
|
+
// to have become right). That is the case the 2026-08-18 note said would
|
|
1164
|
+
// need the FIELD extended rather than the numbers nudged — so
|
|
1165
|
+
// `peakPricing` windows gained `daysOfWeekUtc` and both entries carry
|
|
1166
|
+
// WEEKDAYS_UTC. Modeling them as daily bills a 2× on ~29 weekend hours a
|
|
1167
|
+
// week that DeepSeek does not charge.
|
|
1038
1168
|
// 2. THE CN-VS-US DEFAULT — settled per model, and they differ. Post-rise
|
|
1039
1169
|
// DeepInfra's US flash rates (0.08/0.18/0.016) are BELOW CN's off-peak
|
|
1040
1170
|
// on both input and output, so Flash defaults US; Pro's US re-host still
|
|
@@ -1100,14 +1230,15 @@ export const MODELS = [
|
|
|
1100
1230
|
},
|
|
1101
1231
|
// The peak-hour 2× surcharge is ON the rate card and LIVE since
|
|
1102
1232
|
// 2026-08-16T16:00Z. Peak = 01:00-04:00 and 06:00-10:00 UTC (Beijing
|
|
1103
|
-
// business hours)
|
|
1104
|
-
// like these were once pre-wired ahead of the card and
|
|
1105
|
-
// peak-window turn for weeks — hence the rule that they
|
|
1106
|
-
// once the provider's own page shows them, which it now
|
|
1233
|
+
// business hours) MONDAY THROUGH FRIDAY, at exactly 2× the off-peak rates
|
|
1234
|
+
// above. Windows like these were once pre-wired ahead of the card and
|
|
1235
|
+
// over-billed every peak-window turn for weeks — hence the rule that they
|
|
1236
|
+
// only exist here once the provider's own page shows them, which it now
|
|
1237
|
+
// does.
|
|
1107
1238
|
peakPricing: {
|
|
1108
1239
|
windows: [
|
|
1109
|
-
{ startMinuteUtc: 60, endMinuteUtc: 240 },
|
|
1110
|
-
{ startMinuteUtc: 360, endMinuteUtc: 600 },
|
|
1240
|
+
{ startMinuteUtc: 60, endMinuteUtc: 240, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1241
|
+
{ startMinuteUtc: 360, endMinuteUtc: 600, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1111
1242
|
],
|
|
1112
1243
|
multiplier: 2,
|
|
1113
1244
|
},
|
|
@@ -1161,13 +1292,13 @@ export const MODELS = [
|
|
|
1161
1292
|
regionPricing: {
|
|
1162
1293
|
us: { inputPricePerMTok: 0.08, outputPricePerMTok: 0.18, cacheReadPricePerMTok: 0.016 },
|
|
1163
1294
|
},
|
|
1164
|
-
// Peak-hour surcharge live since 2026-08-16T16:00Z (see
|
|
1165
|
-
// It applies to the NATIVE CN card only — this model
|
|
1166
|
-
// re-host, which is flat, so most turns never take it.
|
|
1295
|
+
// Peak-hour surcharge live since 2026-08-16T16:00Z, Mon-Fri (see
|
|
1296
|
+
// deepseek-v4-pro). It applies to the NATIVE CN card only — this model
|
|
1297
|
+
// defaults to the US re-host, which is flat, so most turns never take it.
|
|
1167
1298
|
peakPricing: {
|
|
1168
1299
|
windows: [
|
|
1169
|
-
{ startMinuteUtc: 60, endMinuteUtc: 240 },
|
|
1170
|
-
{ startMinuteUtc: 360, endMinuteUtc: 600 },
|
|
1300
|
+
{ startMinuteUtc: 60, endMinuteUtc: 240, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1301
|
+
{ startMinuteUtc: 360, endMinuteUtc: 600, daysOfWeekUtc: WEEKDAYS_UTC },
|
|
1171
1302
|
],
|
|
1172
1303
|
multiplier: 2,
|
|
1173
1304
|
},
|
|
@@ -1665,9 +1796,13 @@ export const MODELS = [
|
|
|
1665
1796
|
// GLM context cache: read ≈0.19× input, no write premium.
|
|
1666
1797
|
cacheReadPricePerMTok: 0.26,
|
|
1667
1798
|
cacheWritePricePerMTok: 1.4,
|
|
1668
|
-
//
|
|
1669
|
-
//
|
|
1670
|
-
//
|
|
1799
|
+
// Pinned to the native host, billing the list card above. DeepInfra has
|
|
1800
|
+
// since STARTED serving zai-org/GLM-5.3 (it 404'd when this was cataloged
|
|
1801
|
+
// on 2026-08-26; live 2026-09-02 at $1.20/$4.00, cache read 0.1× = $0.12,
|
|
1802
|
+
// i.e. below native). Adding 'us' here is not a catalog-only edit — it
|
|
1803
|
+
// needs a matching entry in molecule-dev's region-model-maps.ts and a
|
|
1804
|
+
// regionPricing block, moved together, or dispatch sends the canonical id
|
|
1805
|
+
// and 404s. Left as a human decision.
|
|
1671
1806
|
regions: ['cn'],
|
|
1672
1807
|
// Same base weights as glm-5.2, so the same best-effort estimate — Z.ai
|
|
1673
1808
|
// publishes no cutoff.
|
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