Mistral Voxtral Transcribe vs Smallest.ai Pulse STT

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).

Smallest.ai Pulse STT 52

Model reading of the checklist, with training and retention in place of the least-privilege and injection lines. Plain sk_ keys per organisation, shown once, revoked at once and deletable, with one key per environment advised. The docs say a key grants full account access, and no scope is offered. Short-lived access tokens (wat_, 30 to 900 seconds) cover only inference routes and can't mint tokens (22 of 30). The error reference documents the key as a ?token= query parameter for browser WebSockets, and the token guide passes access tokens as ?api_key= on the STT socket, so 10 off (12 of 30). The terms (21 October 2026) say Smallest won't use user content or input to train or improve its services. The privacy policy (18 September 2026) lists improving the services with machine learning as a use of usage data and doesn't address uploaded audio or text either way (15 of 20). Zero data retention and the x-expire-content header are Enterprise only, or a $1,000 a month HIPAA add-on on pay as you go. Other plans keep content readable in logs and the dashboard, with no retention period published (5 of 15). Usage is tracked per key, request rows sit in a dashboard log viewer, and every response carries x-request-id. No audit log of key actions was found (10 of 15). security.txt is valid to 31 December 2026 and the site footer claims SOC 2, GDPR and HIPAA compliance, while the pricing page lists SOC 2 compliance only on Enterprise. No bug bounty, disclosure policy or public advisories found (10 of 20). Total 52.

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.

Smallest.ai Pulse STT 73

Read as a model. The newest Pulse changelog entry is dated 2 November 2026 (updated Indic benchmarks), and Pulse 2.0 launched within the period (30). Dated Pulse entries on 30 August, 18, 29 and 30 September, 1, 12, 13 and 21 October and 2 November (20). Public changelog, email support and a Discord server, with no evidence read of how they answer (10 of 15). The Python SDK smallestai 5.14.0 dates from 7 October with 12 releases since 9 August, but the npm Node SDK has not been published since May 2025 (8 of 15). The Python SDK repository runs CI and release workflows, but has no licence file (5 of 10). Total 73.

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).

Smallest.ai Pulse STT 71

Closed service with clear terms naming Smallest Inc. under California law (15). The ZDR page is specific about what each mode keeps, but the terms' no-training clause and the privacy policy's machine-learning use of usage data sit side by side, the privacy policy gives no retention periods, and the policy doesn't mention the DPA a footer link points to (15 of 30). A deprecations page gives statuses, dates and migration targets, deprecated routes return a Deprecation: true header, and the 30 September changelog removed 14 European codes from the batch enum with a migration to multi-eu. The terms still let Smallest change or discontinue services at any time without notice (15 of 20). Two regions named (Mumbai and Oregon) with region-pinned hosts, and only Google Analytics and Fullstory named among third parties, with no sub-processor list found (10 of 20). Total 55.

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).

Smallest.ai Pulse STT 66

API reading of the checklist. Word timestamps, utterances, diarisation, emotion and gender are each opt-in, so the default response is the transcript with metadata (18 of 25). There are no list endpoints for transcriptions to page. Output controls are the opt-in fields and finalisation settings on the socket (10 of 20). Errors carry a status and a machine-readable code or error_code (LANGUAGE_NOT_SUPPORTED_BY_MODEL, AudioTooLong and others) with fixes, but the body comes in two shapes, message or error (15 of 20). Batch calls are synchronous with no idempotency key, and whether a retried file is billed again isn't stated (10 of 20). A file call needs only the model and audio. Official Python and Node SDKs exist, though the Node SDK was last published in May 2025 (13 of 15). Total 66.

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).

Smallest.ai Pulse STT 63

BetterStack status page at status.smallest.ai with components for the Global, India and USA APIs (one health check covers the Agents and Models APIs), docs, the app and Electron, and per-component history (20). In the 90 days from 8 August the health checks on the shared API failed in all three regions on 9, 10, 13, 14 and 15 September (about 8, 21, 33, 17 and 64 minutes of flapping), in the Global and India regions on 11 and 17 September (12 minutes each), and in the USA region alone on 1, 3, 5, 12, 21 and 30 September (6 to 62 minutes) and 6 October (12 minutes). Outages in every region at once are read as wide errors, so several majors (0). Standard plan limits published per model, 100 concurrent streaming sessions and 25 batch requests a minute, and the pricing page also says 100 concurrent STT streams (15). 429 is documented with exponential backoff from 500 ms to 30 s and Retry-After when present, and batch calls are synchronous with optional webhooks (15). An infrastructure SLA of 99.99 per cent is named for Enterprise only, with no terms published, and the terms disclaim any availability commitment (5 of 10). Pulse, Pulse 2.0 and Pulse Pro carry no preview label, but eleven language codes are marked Beta (8 of 10). 20 + 0 + 15 + 15 + 5 + 8 = 63.

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).

Smallest.ai Pulse STT 92

OpenAPI 3.1 for the HTTP routes and AsyncAPI 2.6 for the STT socket, with query parameters and event schemas (25). llms.txt and a Markdown copy of every page (10). Model cards and a compare page say which model suits which audio, and fields say where they're ignored (Pulse Pro ignores redaction), but several changelog entries correct response fields the reference had wrong, such as the diarisation shape and a redacted_entities array that doesn't exist (15 of 20). Typed query parameters with enums, defaults and required marks, though booleans are string enums that must be lowercase (13 of 15). Examples in curl, Python and JavaScript, and the reference lists 400, 401, 403, 413, 422 and 429 bodies with known code values (14 of 15). Versioned /v1 paths, a deprecations page and a dated Pulse changelog (15). Total 92.