@grida/ai-models 1.20261111.0 → 1.20261111.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/PROVENANCE.md CHANGED
@@ -600,13 +600,201 @@ Source: replicate.com/google/lyria-3 — "$0.04 per output audio file"
600
600
 
601
601
  Source: replicate.com/google/lyria-3-pro — "$0.08 per output audio file"
602
602
 
603
- ## `pricing: {`
604
-
605
- API: 100 credits when duration is automatic, or 11 credits/s when set.
606
-
607
- ## `pricing: { type: "per_1000_characters", usd: 0.1 },`
608
-
609
- Source: elevenlabs.io/pricing/api — $0.10 per 1,000 characters.
603
+ ## ElevenLabs model family and serving contracts
604
+
605
+ Verified on 2026-11-11 against the native
606
+ [model overview](https://elevenlabs.io/docs/overview/models),
607
+ [Vercel catalogue](https://ai-gateway.vercel.sh/v1/models),
608
+ [OpenRouter complete catalogue](https://openrouter.ai/api/v1/models?output_modalities=all)
609
+ and [fal ElevenLabs results](https://fal.ai/api/models?keywords=elevenlabs&size=100).
610
+ The cards cover the native model table, the still-available named legacy models,
611
+ and every ElevenLabs model in the Vercel and OpenRouter feeds. Native underscore
612
+ IDs and the existing `eleven_v3` / `eleven_text_to_sound_v2` identities are retained;
613
+ each provider keeps its own exact call ID. Voice design, voice conversion,
614
+ transcription and conversational voice agents are separate audio families.
615
+
616
+ ### Compatibility and native contracts
617
+
618
+ Shared discovery metadata uses the Vercel Gateway `type` and `modalities`
619
+ vocabulary. The audio sections organize entries; they do not introduce new
620
+ request or event schemas. Native voice design and voice conversion have media
621
+ modalities but no invented baseline type. Vercel bindings retain the exact
622
+ `type` and `modalities` values from its published feed.
623
+
624
+ Provider-specific language coverage, character limits, commit strategies,
625
+ timestamps and format choices stay in their native binding's qualifications or
626
+ authoritative contract. New audio cards do not define a shared `transcription`
627
+ capability object, a display-language field or a copied output-format enum.
628
+ The two pre-existing cards retain their released v1 direct-provider fields,
629
+ with the sourced corrections below.
630
+
631
+ Request and result schema URLs point directly into the official
632
+ [ElevenLabs OpenAPI 3.1 document](https://api.elevenlabs.io/openapi.json),
633
+ [OpenRouter OpenAPI 3.1 document](https://openrouter.ai/openapi.json), and fal's
634
+ per-model queue OpenAPI documents, for example
635
+ [v4](https://fal.ai/api/openapi/queue/openapi.json?endpoint_id=elevenlabs%2Ftts%2Feleven-v4).
636
+ The JSON Pointer targets were resolved against the actual published documents.
637
+ Native schema keywords, enums and dialects are not rewritten as partial custom
638
+ parameter metadata. fal's result schema links describe result retrieval, not
639
+ the QueueStatus returned by queue submission.
640
+
641
+ The official native
642
+ [transcription](https://elevenlabs.io/docs/api-reference/speech-to-text/v-1-speech-to-text-realtime)
643
+ and [dialogue WebSocket](https://elevenlabs.io/docs/api-reference/text-to-dialogue/ttd-websocket)
644
+ references publish their AsyncAPI contracts. Streaming messages remain defined
645
+ there; the catalogue does not invent a shared event protocol. Voice conversion
646
+ is described in the native reference but is absent from the public OpenAPI
647
+ document inspected, so no machine-readable schema URL is guessed for that route.
648
+
649
+ ### Native prices and output corrections
650
+
651
+ The [native API pricing page](https://elevenlabs.io/pricing/api) now publishes USD
652
+ prices rather than the old sound-effects credit counts:
653
+
654
+ - v4, v3 and Multilingual v2: **$0.08 per 1,000 characters**. v4 Turbo, v3
655
+ Conversational, Flash and Turbo: **$0.04 per 1,000 characters**. The native
656
+ `eleven_v3` card's former $0.10 rate is corrected; fal's independent $0.10 rate
657
+ remains unchanged.
658
+ - Sound Effects: **$0.12 per minute**, normalized exactly to **$0.002 per second**.
659
+ Music: **$0.15 per minute**, normalized exactly to **$0.0025 per second**.
660
+ These amounts do not establish whole-minute rounding or a minimum charge.
661
+ - Scribe v3, v2 and v2 Medical: **$0.20 per input hour**. Scribe v3 Live and v2
662
+ Realtime: **$0.22 per input hour**. Optional keyterms, transcript editing and
663
+ the v3 Live keyterm library add charges. The batch API's percentage surcharge
664
+ descriptions disagree with the price table's hourly add-ons, so only the
665
+ undisputed base rates are recorded. The table does not name legacy Scribe v1;
666
+ its version-specific native price remains unknown.
667
+ - Voice Changer: **$0.12 per input minute**. Voice Design is metered by preview
668
+ text characters, but the [voice-design guide](https://elevenlabs.io/docs/eleven-creative/voices/voice-design)
669
+ does not establish a version-specific regular USD rate; none is invented.
670
+ - Eleven Realtime: standard connected audio **$0.08/minute**, or **$0.16/minute**
671
+ above the published concurrency allowance, plus token meters. Mini input/output
672
+ is **$0.20/$6 per million tokens**; Max is **$1.80/$9**. Token prices remain
673
+ structured in `token_pricing`; connected-audio rates and the concurrency
674
+ condition remain sourced qualifications. No synthetic request parameter is
675
+ introduced to express session/account concurrency context, and the token
676
+ amounts alone are not the total session price.
677
+
678
+ The native [TTS](https://elevenlabs.io/docs/api-reference/text-to-speech/convert),
679
+ [dialogue](https://elevenlabs.io/docs/api-reference/text-to-dialogue/convert),
680
+ [sound-effects](https://elevenlabs.io/docs/api-reference/text-to-sound-effects/convert)
681
+ and [voice-conversion](https://elevenlabs.io/docs/api-reference/speech-to-speech/convert)
682
+ references publish selectable MP3, PCM, WAV, Opus, mu-law and A-law formats.
683
+ The existing v3 and Sound Effects v2 cards therefore no longer imply MP3-only
684
+ output. Sample rate and bitrate describe the default output, not every selectable
685
+ format; some high-quality combinations require particular plans.
686
+
687
+ HTTP Text to Dialogue accepts up to ten voices and recommends at most 2,000
688
+ total text characters for reliable generation. This is separate from the model's
689
+ single-speaker character envelope. Conversational variants reference the
690
+ [dialogue WebSocket](https://elevenlabs.io/docs/eleven-api/guides/how-to/websockets/realtime-tdd),
691
+ whose guide demonstrates MP3 output; the HTTP format list is not silently copied
692
+ to it. The [music API](https://elevenlabs.io/docs/api-reference/music/compose)
693
+ defaults to 44.1 kHz / 128 kbps for v1 and 48 kHz / 192 kbps for v2 models.
694
+ Its 3–600 second `music_length_ms` range applies to prompt-based generation,
695
+ while v2 composition plans are chunk-based and enforce section durations.
696
+
697
+ ### Provider-specific prices, limits and identity gaps
698
+
699
+ - Vercel and OpenRouter publish **$80/M characters** for v4, v3 and Multilingual
700
+ v2, and **$40/M characters** for their lower-latency variants. OpenRouter's
701
+ per-model endpoint metadata, for example
702
+ [v3](https://openrouter.ai/api/v1/models/elevenlabs/eleven-v3/endpoints),
703
+ confirms `pricing.discount: 0` for every added binding. Character units come
704
+ from each model page, not the generic `prompt` field name or a launch banner.
705
+ - Vercel's feed records Scribe v2 at exactly **$0.000061112 per input second**;
706
+ retain that decimal rather than replacing it with the rounded $0.22/hour
707
+ display. OpenRouter's Scribe v2 and Medical endpoint metadata records exactly
708
+ **$0.0000611111111111 per input second**. Neither is a per-token price.
709
+ - The Vercel feed leaves Scribe v2 Realtime pricing empty, but its
710
+ [model page](https://vercel.com/ai-gateway/models/scribe-v2-realtime) publishes
711
+ **$0.39 per hour of streamed audio**. The page supplies that price; an empty
712
+ feed object does not mean free inference.
713
+ - Vercel's Music v2/v2.5 and Sound Effects v2 output-duration meters are
714
+ **$0.0025/s** and **$0.002/s** respectively. Its Realtime Mini/Max bindings add
715
+ **$0.0013334 per connected session second** to separate input, cached-input
716
+ and output token meters. Preserve the exact feed amounts and separate meters.
717
+ - fal [v4](https://fal.ai/models/elevenlabs/tts/eleven-v4/api) and
718
+ [v4 Turbo](https://fal.ai/models/elevenlabs/tts/eleven-v4-turbo/api) accept only
719
+ **5,000 text characters** per request. Other providers' 10,000-character
720
+ envelopes must not override these serving limits. fal's old
721
+ [Sound Effects v2 schema](https://fal.ai/models/fal-ai/elevenlabs/sound-effects/v2/api)
722
+ restricts durations to **0.5–22 seconds**, while the native API and fal's v3
723
+ endpoint support up to 30 seconds. Both fal SFX schemas cap text at 450 characters.
724
+ - fal [Music v2](https://fal.ai/models/elevenlabs/music/v2) and
725
+ [v2.5](https://fal.ai/models/elevenlabs/music/v2.5) cost **$0.60 per output
726
+ minute, rounded up to the next whole minute**. A 30-second output therefore
727
+ costs $0.60. This explicit rounding is preserved in `published_pricing`.
728
+ - fal's named [Scribe v1](https://fal.ai/models/fal-ai/elevenlabs/speech-to-text)
729
+ costs **$0.03/minute**; its
730
+ [Scribe v2](https://fal.ai/models/fal-ai/elevenlabs/speech-to-text/scribe-v2)
731
+ costs **$0.008/minute**, with a 30% surcharge when using keyterms. The v3
732
+ [multi-speaker dialogue endpoint](https://fal.ai/models/fal-ai/elevenlabs/text-to-dialogue/eleven-v3)
733
+ is retained beside single-speaker TTS at its separately verified $0.10/1,000
734
+ characters, with its different input contract qualified.
735
+ - fal's unversioned [voice changer](https://fal.ai/models/fal-ai/elevenlabs/voice-changer/api)
736
+ and [music endpoint](https://fal.ai/models/fal-ai/elevenlabs/music/api) do not
737
+ identify an exact native version or expose a model selector. The affected
738
+ versioned cards record identity exclusions rather than guessing a binding.
739
+ Utilities such as audio isolation, alignment and dubbing are not assigned
740
+ invented native model IDs.
741
+
742
+ OpenRouter's [Models API documentation](https://openrouter.ai/docs/api/api-reference/models/list-all-models-and-their-properties)
743
+ says the unfiltered endpoint defaults to text-output models. Audio absence
744
+ checks use `output_modalities=all`; Sound Effects v2 remains absent, while the
745
+ nine listed TTS models and two Scribe models have explicit bindings.
746
+
747
+ The native model overview contains conflicting Scribe v3 Live keyterm-library
748
+ limits: the feature summary says 100,000 terms, but the detailed section and
749
+ [dedicated keyterm-library guide](https://elevenlabs.io/docs/eleven-api/guides/how-to/speech-to-text/batch/keyterm-prompting)
750
+ specify 10,000. The card uses the more specific 10,000-term contract and qualifies
751
+ the conflict. Similarly, the current model reference and November 2 changelog
752
+ explicitly support SFX v3 looping, while the developer launch article still says
753
+ v2-only. The explicit current reference and changelog are used for looping.
754
+ The [voice-changer guide](https://elevenlabs.io/docs/overview/capabilities/voice-changer)
755
+ limits input segments to five minutes; this audio limit is recorded separately
756
+ from the unrelated character limits in the model overview's TTS table.
757
+
758
+ ### Release provenance
759
+
760
+ | Models | First broad availability | First-party evidence |
761
+ | -------------------------- | ------------------------ | ---------------------------------------------------------------------------------------- |
762
+ | Multilingual v2 | 2023-08-22 | [Announcement](https://elevenlabs.io/blog/eleven-multilingual-v2) |
763
+ | Turbo v2 | 2023-11-09 | [Announcement](https://elevenlabs.io/blog/turbo-v2-is-here) |
764
+ | Turbo v2.5 | 2024-07-19 | [Announcement](https://elevenlabs.io/blog/introducing-turbo-v25) |
765
+ | Flash v2 / v2.5 | 2024-12-18 | [Announcement](https://elevenlabs.io/blog/meet-flash) |
766
+ | Scribe v1 | 2025-02-26 | [Announcement](https://elevenlabs.io/blog/meet-scribe) |
767
+ | Eleven v3 | 2025-06-03 | [Public preview](https://elevenlabs.io/blog/eleven-v3) |
768
+ | Voice Design v3 | 2025-06-25 | [Announcement](https://elevenlabs.io/blog/voice-design-v3) |
769
+ | Music v1 | 2025-08-05 | [Announcement](https://elevenlabs.io/blog/eleven-music-is-here) |
770
+ | Scribe v2 Realtime | 2025-11-11 | [Announcement](https://elevenlabs.io/blog/introducing-scribe-v2-realtime) |
771
+ | Scribe v2 | 2026-01-09 | [Announcement](https://elevenlabs.io/blog/introducing-scribe-v2) |
772
+ | Eleven v3 Conversational | 2026-02-09 | [Named agent model addition](https://elevenlabs.io/docs/changelog/2026/2/9) |
773
+ | Music v2 | 2026-05-26 | [Public product launch](https://elevenlabs.io/blog/introducing-music-v2) |
774
+ | Music v2.5 | 2026-09-11 | [Announcement](https://elevenlabs.io/blog/music-v2-5-model) |
775
+ | Scribe v2 Medical | 2026-09-22 | [General availability](https://elevenlabs.io/blog/scribe-v2-medical-generally-available) |
776
+ | Eleven v4 / v4 Turbo | 2026-09-28 | [Announcement](https://elevenlabs.io/blog/eleven-v4) |
777
+ | Eleven Realtime Mini / Max | 2026-11-02 | [Broad availability](https://elevenlabs.io/docs/changelog/2026/11/2) |
778
+ | Sound Effects v3 | 2026-11-02 | [Broad availability](https://elevenlabs.io/docs/changelog/2026/11/2) |
779
+ | Scribe v3 / v3 Live | 2026-11-10 | [Announcement](https://elevenlabs.io/blog/scribe-v3) |
780
+
781
+ Music v2's public product launch precedes its
782
+ [2026-06-15 API availability](https://elevenlabs.io/docs/changelog/2026/6/15).
783
+ Music v2's earlier public product date is used. Realtime and Sound Effects v3
784
+ require a different distinction: the
785
+ [October 19 changelog](https://elevenlabs.io/docs/changelog/2026/10/19) describes
786
+ a gradual Realtime rollout, while the
787
+ [November 2 changelog](https://elevenlabs.io/docs/changelog/2026/11/2) explicitly
788
+ opens both Realtime models and SFX v3 to every workspace. The earlier Realtime
789
+ and SFX announcements do not prove broad access during that rollout, so their
790
+ release dates use November 2. Provider onboarding dates and opaque `created`
791
+ timestamps are not substituted for model availability.
792
+
793
+ No exact variant launch day was established for `eleven_multilingual_ttv_v2`,
794
+ `eleven_multilingual_sts_v2` or `eleven_english_sts_v2`. Their documented native
795
+ endpoints carry `release.date: null` and `basis: "provider_endpoint"`; a generic
796
+ Voice Design or Voice Changer product launch does not prove the exact v2 date.
797
+ The existing Sound Effects v2 release provenance is preserved.
610
798
 
611
799
  ## `"tripo/rig-v2.5": {`
612
800
 
package/README.md CHANGED
@@ -5,7 +5,7 @@ provider bindings, and published prices, with the labs that make them and their
5
5
  logos. Every fact cites a first-party source.
6
6
 
7
7
  This package is the data, published unchanged from release
8
- [`v1.20261111.0`](https://github.com/gridaco/models/releases/tag/v1.20261111.0) of [gridaco/models](https://github.com/gridaco/models). It contains no code.
8
+ [`v1.20261111.1`](https://github.com/gridaco/models/releases/tag/v1.20261111.1) of [gridaco/models](https://github.com/gridaco/models). It contains no code.
9
9
 
10
10
  | file | contents |
11
11
  | --- | --- |
@@ -25,7 +25,7 @@ From a CDN, without installing:
25
25
 
26
26
  ```
27
27
  https://cdn.jsdelivr.net/npm/@grida/ai-models@1/models.json
28
- https://cdn.jsdelivr.net/npm/@grida/ai-models@1.20261111.0/models.json
28
+ https://cdn.jsdelivr.net/npm/@grida/ai-models@1.20261111.1/models.json
29
29
  ```
30
30
 
31
31
  ## Versions
package/manifest.json CHANGED
@@ -1,19 +1,19 @@
1
1
  {
2
2
  "name": "grida-models",
3
- "version": "v1.20261111.0",
3
+ "version": "v1.20261111.1",
4
4
  "schema_version": 1,
5
- "commit": "a214c918873e261463adc6f809f6783726223261",
6
- "created": "2026-11-11T13:26:04Z",
7
- "source": "https://github.com/gridaco/models/tree/a214c918873e261463adc6f809f6783726223261/catalogue",
8
- "digest": "sha256:d6edf490dfb879632cff1501e32d963156137a8f6b9c2b0f1ccad9a9763228c4",
5
+ "commit": "86280ed1c1630db858aad8a234c02f9e6f371e64",
6
+ "created": "2026-11-11T13:27:11Z",
7
+ "source": "https://github.com/gridaco/models/tree/86280ed1c1630db858aad8a234c02f9e6f371e64/catalogue",
8
+ "digest": "sha256:061ca29bd47e1ef3f87b6efc527394c41901f41e62f54007e7ef199f11544988",
9
9
  "counts": {
10
- "cards": 81,
10
+ "cards": 105,
11
11
  "labs": 18
12
12
  },
13
13
  "files": {
14
14
  "PROVENANCE.md": {
15
- "sha256": "d55006401d872e875de018586cdd2d4ec3eb4b839f467c51760f500bb80e7b58",
16
- "bytes": 83197
15
+ "sha256": "702dc3431e80ef3b5fca1f0f033f4018c8c9c1abdfb7919ebd3b5f74b8eb726a",
16
+ "bytes": 97528
17
17
  },
18
18
  "README.md": {
19
19
  "sha256": "ffb53916698e78b394c8440edc1b3ea5b70e3e31cf67a734bf89a469c18ed17f",
@@ -92,8 +92,8 @@
92
92
  "bytes": 330
93
93
  },
94
94
  "models.json": {
95
- "sha256": "ecddf1afd7cbf24ed09f27391bb659fa863f953f81ea271aef408fc3f3db2637",
96
- "bytes": 822701
95
+ "sha256": "471361b833653d2f702da168d4177aa737f97ddc7bb2fa93b179dff70c0d6249",
96
+ "bytes": 957100
97
97
  },
98
98
  "schemas/labs.schema.json": {
99
99
  "sha256": "9ddaedd56826bebdbe72b84f76cfc86df0b5c5902ba9f0259c2ea2cbb9327f99",