Inworld STT vs Mistral Voxtral Transcribe

Inworld STT 41

Read as a model. No public STT changelog. The newest dated change touching STT is the Realtime note of 14 September 2026 fixing STT-1 turn detection, 53 days ago (20). The CLI, which carries the STT command and MCP tool, shipped 1.4.0 to 1.5.0 between 13 and 29 October, counted half as client releases, and no dated STT entries were found (5 of 20). Discord and support@inworld.ai, with no evidence of answers read (6 of 15). No official STT SDK; the CLI is current (5 of 15). CLI on npm under MIT, current, CI not read (5 of 10). Total 41.

Mistral Voxtral Transcribe 88

Model reading. Mistral released three Voxtral 3 transcription models (Mini Transcribe 3, Small Transcribe 3 and Mini Transcribe Realtime 3) in Public Preview on 20 October 2026, 13 days before this check (30). The lifecycle policy promises six months' notice for GA models, and Transcribe 2 and Realtime 2602 were deprecated on 31 October with retirement on 30 April 2027, 181 days (12 of 12). No transcription model was retired in the 90 days to 2 November; Voxtral Mini Transcribe (2507) retired on 31 May 2026 (8 of 8). Six dated changelog entries from 31 August to 22 October, one of them the Voxtral 3 release, read as mistral-api reads the same changelog (10 of 15). The eight newest open issues on client-python had zero or one comment per the 8 October check (5 of 10). Official mistralai 3.7.0 (1 November) per the client-python RELEASES.md, and @mistralai/mistralai, with a v2 to v3 migration guide (15). Both are generated from the OpenAPI document, with test, lint and example workflows in the repository (8 of 10). 30+12+8+10+5+15+8 = 88.

Inworld STT 54

No OpenAPI or other machine-readable file. The usual spec paths redirect, and the reference pages are Markdown (0). llms.txt, Markdown twins and a docs MCP server (10). Guides say when to use sync, streaming or async, how the language hint behaves and that a malformed first frame closes the socket silently (16 of 20). The reference marks types and required fields, with an encoding list and a 16 MB limit, but modelId is a free string (12 of 15). Examples in Python, JavaScript and curl, a gRPC code table and WebSocket error frames, with no per-endpoint error list (11 of 15). Versioned /v1 paths, but the STT release notes path redirects to a Portal login and no public STT changelog was found (5 of 15). Total 54.

Mistral Voxtral Transcribe 85

Model reading. The public OpenAPI document covers /v1/audio/transcriptions, its SSE streaming form and /v1/client/sessions. The realtime WebSocket messages are described by SDK types and guide examples, not by the document (22 of 25). llms.txt with Markdown twins of the speech-to-text, offline, realtime and client authentication pages (10). The guides say which model each endpoint runs, that timestamp_granularities and language cannot be combined and that realtime cannot diarise, and carry an OpenAI SDK compatibility table. In the OpenAPI document diarize, context_bias and temperature have no description (14 of 20). timestamp_granularities is an enum of segment and word, language is a two-letter pattern and context_bias items have a pattern, while model is a free string (12 of 15). curl, Python and TypeScript examples on each guide and a shared error glossary. The OpenAPI response example still names voxtral-mini-2507 (12 of 15). Dated model identifiers and a dated changelog (15).

Inworld STT 60

Closed service with clear terms naming Theai, Inc. dba Inworld AI (15). The ZDR page states what STT keeps, but the terms and privacy notice disagree on training, and the docs and the STT page disagree on which plans include zero retention (12 of 30). No dated STT deprecation or change notices are public (8 of 20). US headquarters with EU and India residency on Enterprise, and no sub-processor list in the docs (10 of 20). Total 45.

Mistral Voxtral Transcribe 83

The hosted service is closed under commercial terms. The Realtime weights are published on Hugging Face under Apache-2.0 and the SDKs are Apache-2.0, while Voxtral Mini Transcribe 2 has no public weights (20 of 30). The privacy policy effective 3 September 2026 (30 rolling days), the zero data retention page and the terms agree on retention, and the terms leave the API's training default unnamed (20 of 30). A model lifecycle page with notice periods per stage (20). Opt-in EU and US regional endpoints are documented, the global endpoint commits to no location, and the provider list is in a trust centre we could not read (13 of 20).

Inworld STT 56

API reading of the checklist. Voice Profile, word timestamps and diarisation are opt-in, and responses carry a usage block (18 of 25). No lists to page. Async jobs return an operation to poll (8 of 20). Errors carry a numeric gRPC code and message on both endpoints, but a malformed first message closes the socket with 1005 and no frame (12 of 20). Sync calls are stateless and async calls return an operation, with no idempotency key, and empty transcripts and add-ons are billed (10 of 20). A call needs a model, an encoding and the audio. No STT SDK was found, only the CLI and its MCP server and REST examples (8 of 15). Total 56.

Mistral Voxtral Transcribe 77

API reading, as for the other speech-to-text listings. Segments, word timestamps and speaker labels are returned only when asked for, and the default response is the text, the language and usage. No subtitle or plain-text output format was found (18 of 25). There is no transcript store to page through. Volume goes through the batch API, which accepts /v1/audio/transcriptions (12 of 20). Errors return type, param and code with a fix per status in the glossary (17 of 20). The call is stateless and safe to retry, the SDKs retry with backoff, and no idempotency key exists or is needed (15 of 20). model is the only required field beside the audio source, and there are official Python and TypeScript SDKs plus OpenAI SDK compatibility for the batch endpoint (15).

Inworld STT 60

Hosted reading of the checklist. status.inworld.ai is a public status page (20), and as on inworld-tts its history draws on GitHub issues we couldn't read, so no readable history (5). Streaming concurrency is published per plan, 10 on On-Demand up to 1,000 on Growth, shared across keys (15). 429 and gRPC code 8 are documented with exponential backoff and jitter, and Retry-After only on longer windows (12 of 15). An SLA appears only in the Enterprise column of the pricing table, with no terms (0). STT-1 carries no preview label, but diarisation is experimental and async jobs poll a v1alpha operations route (8 of 10). 20 + 5 + 15 + 12 + 0 + 8 = 60.

Mistral Voxtral Transcribe 53

Hosted reading. The docs point to status.mistral.ai, a Rootly page with component uptime bars per the 1 and 2 October reading on the other Mistral listings (20). The page has answered with a Cloudflare bot check since 8 October, robots.txt included on 2 November, so it is treated as blocked and the 90-day record for audio was not read (5, unread and not a finding against the vendor). The limits page names audio seconds a minute and audio seconds a month as the audio limits, with the numbers shown only in the Admin Panel, per the 8 October check (3 of 15). The error glossary tells callers to back off exponentially on 429, 500, 502, 503 and 504 and to read the Retry-After header, and the official SDKs retry with backoff. Transcription is a stateless call (15). No SLA found (0). Voxtral Mini Transcribe 2 and Realtime (2602) are deprecated but served until 30 April 2027, and voxtral-mini-latest still points to Transcribe 2, while the Voxtral 3 transcription models are Public Preview (10).

Inworld STT 63

Model reading of the checklist, with training and retention in place of the least-privilege and injection lines. Workspace API keys sent as Basic credentials, deleted to revoke, with separate keys per environment advised. The CLI mints read-only keys unless --write is given, but write toggles cover only the Voice and Router APIs, so no key is limited to one product. Browsers use short-lived tokens minted server side, via the bearer_ WebSocket subprotocol, not the URL (25 of 30). The Terms (11 June 2025) say Inworld won't train generally available models on customers' non-public Materials, except feedback and flagged content, while the Privacy Notice (18 January 2026) lists training AI models among its uses, per the inworld-tts check of 4 November (12 of 20). Zero data retention keeps STT audio and transcripts in memory and out of logs, included on Developer ($300 a month) and above per the docs, while the STT page says Enterprise or a Growth add-on. Async results and audio go after 7 days, and no default period is given for logs (8 of 15). Usage is attributed per key and a Performance page shows errors and throttling (8 of 15). SOC 2 Type II and HIPAA claimed and a vulnerability report form, with no security.txt or bug bounty, per the 4 November reading (10 of 20). Total 63.

Mistral Voxtral Transcribe 70

Model reading, scored like the other Mistral listings. Keys are bound to one workspace, can carry an expiry date and a connector scope, and organisations can enforce a maximum key lifetime. Browser clients get rt_ tokens from POST /v1/client/sessions that last about 900 seconds, work for one model and travel in Sec-WebSocket-Protocol. A key cannot be limited to transcription (26 of 30). The commercial terms effective 25 September 2026 say Mistral will not train on customer data unless the customer opted in, or did not opt out where a product defaults to opt-in, without naming the API's default. The privacy and data controls page the docs link to returned 404 (10 of 20). Input and output are kept 30 rolling days for abuse monitoring, and /v1/audio/transcriptions is on the zero data retention list for paid plans, on request and subject to approval (12 of 15). Audit logs cover API key actions on Enterprise plans only (10 of 15). security.txt is valid until 5 May 2027 and names a HackerOne submission form, and the docs publish three security advisories for 2026. Certifications sit in a trust centre drawn by script that we could not read (12 of 20).