Mistral Voxtral Transcribe

# Fix list: Mistral Voxtral Transcribe

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/mistral-voxtral-transcribe, the October 2026 research run, assessed 8 October 2026. Grade B, 64.1 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Mistral Voxtral Transcribe: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 53 out of 100, up to 9.4 more on the total

Why it scored 53: Hosted reading. The docs point to status.mistral.ai, which our sibling Mistral listings read on 1 and 2 October 2026 as a Rootly page with component uptime bars (20). On 8 October the page answered curl and WebFetch with a Cloudflare check and HTTP 403, so 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 (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). The model cards mark Voxtral Mini Transcribe 2 and Voxtral Mini Transcribe Realtime as GA (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0). $0.003 a minute for Voxtral Mini Transcribe 2 and $0.006 a minute for Voxtral Realtime on public pages (20). Free mode gives API access with no card, within limits shown in the console (20). A person signs up in a browser (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 3. Security & auth, 70 out of 100, up to 5.3 more on the total

Why it scored 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).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 4. Agent ergonomics, 77 out of 100, up to 3.7 more on the total

Why it scored 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).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Maintenance & community, 66 out of 100, up to 3 more on the total

Why it scored 66: Model reading. The last dated change to transcription is the 4 February 2026 release of both models with diarisation and context biasing, eight months ago. The SDKs that carry the realtime client were released on 6 October 2026 (Python 3.1.0), so we give 10 of 30, a departure from the strict reading of 0. The lifecycle policy promises six months' notice for GA models (12 of 12). No transcription model was retired in the last 90 days of the changelog (8 of 8). Five dated platform changelog entries between 16 July and 6 October, none about audio (8 of 15). The eight newest open issues on client-python have zero or one comment (5 of 10). Official `mistralai` 3.1.0 on PyPI and `@mistralai/mistralai` 2.7.0 on npm (15). Both are generated from the OpenAPI document, with test, lint and example workflows in the repository (8 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## 6. Schema & documentation, 85 out of 100, up to 2.4 more on the total

Why it scored 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).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Transparency & trust, 82 out of 100, up to 1.6 more on the total

Made of editorial 73, provenance 91.

Why it scored 82: 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).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: mistral.ai, registered 2019-05-15 (7 years) (11 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 2 clauses that cost points (5.1 of 10)

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 2026-05-12. Compromised `mistralai` 2.4.6 on PyPI ran a credential-harvesting script on import for about three hours, and three `@mistralai/mistralai` versions on npm were also replaced. Mistral published advisory MAI-2026-002, removed the packages and closed its investigation on 14 May. Fixed and documented, so 3 points. https://docs.mistral.ai/resources/security-advisories/MAI-2026-002

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: status.mistral.ai answered every request with a Cloudflare check (HTTP 403) on 8 October 2026, so the audio component's uptime and the 90-day incident record were not read
- unchecked: trust.mistral.ai is drawn by script, so certifications and the provider list were not read
- unchecked: the privacy and data controls page linked from the zero data retention page returned 404, so the API's training default was not established
- unchecked: whether Free mode includes transcription minutes and whether sign-up needs a phone number. The quickstart says no card is needed
- The pricing table shows $0.0003 a minute in a second column for Voxtral Mini Transcribe 2 without a label we could tie to batch or cached use, so it is not listed as a price
- Whether zero data retention covers the realtime WebSocket at `/v1/audio/transcriptions/realtime`. The list names `/v1/audio/transcriptions` only
- The transcription curl examples send the key in an `x-api-key` header, while the error glossary gives `Authorization: Bearer`
- The other Mistral listings took no deduction for advisory MAI-2026-002. This one takes 3 because the realtime client is only reachable through the SDK

## Weaknesses

- Audio rate limits are named (audio seconds a minute and a month) but the numbers are shown only in the Admin Panel
- `timestamp_granularities` cannot be used with `language`, and the realtime model cannot diarise
- 13 languages, and context biasing is tuned for English with other languages described as experimental
- The compromised `mistralai` 2.4.6 on PyPI harvested credentials on import for three hours on 12 May 2026
- No SLA found, and status.mistral.ai answered our requests with a bot check on 8 October 2026
- The Studio MCP server has no transcription tool

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Send `model=voxtral-mini-latest` and one of `file`, `file_url` or `file_id` as multipart form fields to `/v1/audio/transcriptions`
- Leave `language` out when you ask for `timestamp_granularities`, the docs say the two are not compatible
- Use the batch endpoint for diarisation. `voxtral-mini-transcribe-realtime-2602` does not accept `diarize`
- Pin `voxtral-mini-2602` if output must not change, since `-latest` aliases can move
- Mint browser tokens with `POST /v1/client/sessions` close to connection time and pass them in `Sec-WebSocket-Protocol`, never the API key

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.