Groq Speech-to-Text

# Fix list: Groq Speech-to-Text

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/groq-speech-to-text, the October 2026 research run, assessed 8 October 2026. Grade BB, 71.8 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 Groq Speech-to-Text: 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. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0). $0.04 an audio hour for Whisper Large v3 Turbo and $0.111 for Whisper Large v3 on the models page and model cards, read without a login (20). The free plan needs no card and includes both Whisper models at its own rate limits (20). A person signs up in the console 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.

## 2. Schema & documentation, 58 out of 100, up to 6.8 more on the total

Why it scored 58: Model reading. No OpenAPI document was found in the docs, and the Python SDK's `.stats.yml` carries an endpoint count with no specification URL (0). llms.txt at console.groq.com with Markdown twins of the guide, model cards and reference (10). The guide says which model to pick for accuracy or price, that translation is English only and that the `prompt` steers style within 224 tokens, and it explains the `verbose_json` quality fields (15 of 20). The reference lists allowed values for `response_format` and `timestamp_granularities[]` and marks `model` as required, while `language` and `url` are plain strings (11 of 15). curl, Python and JavaScript examples for both endpoints and a shared errors page with a typed error object. The Batch page's sample still names `distil-whisper-large-v3-en`, shut down on 23 August 2025 (12 of 15). The docs changelog is dated, its newest entry is 18 April and no entry in it concerns speech-to-text (10 of 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.

## 3. Agent ergonomics, 75 out of 100, up to 4.1 more on the total

Why it scored 75: API reading, as for the other speech-to-text listings. The default response is the text alone, `text` returns a bare string, and segments and word timestamps come only with `verbose_json`. `srt` and `vtt` are listed as unsupported (18 of 25). There is no transcript store to page through. Volume goes through the Batch API by `url`, and a file over 25 MB (free) or 100 MB (Developer) has to be chunked by the caller (12 of 20). The errors page gives 13 status codes with a recovery line each and an error object of `message` and `type`, with no machine-readable code field (15 of 20). The call is stateless and safe to retry, and the SDKs retry connection errors, 408, 409, 429 and 5xx by default. No idempotency key exists or is needed (15 of 20). `model` and one of `file` or `url` are the only required fields, with official Python and TypeScript SDKs and OpenAI SDK compatibility (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.

## 4. Security & auth, 79 out of 100, up to 3.7 more on the total

Why it scored 79: Model reading, scored like the GroqCloud listing. Bearer keys belong to one project, each project has its own rate limits per model, and the security onboarding page covers rotation and immediate revocation. No way to limit a key to the audio endpoints was found (26 of 30). The services agreement, last modified 22 June 2026, says Groq may not use inputs or outputs for training or fine-tuning unless the customer grants permission (20). Inputs and outputs are not retained by default, logs kept for reliability or abuse last up to 30 days, and the data page lists both audio endpoints as eligible for zero data retention, a setting in Data Controls (15). Usage data and request logs per project and a read-only Reader role. No audit log of administrative actions was found (10 of 15). security.txt on groq.com holds a Contact line only, with no expiry date. The DPA commits to SOC 2 Type II audits, and the trust centre at trust.groq.com is drawn by script and was not read (8 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.

## 5. Maintenance & community, 68 out of 100, up to 2.8 more on the total

Why it scored 68: Model reading, as for the GroqCloud listing. No dated change to speech-to-text was found in 2026. The newest dated release is the Python SDK 1.7.0, uploaded to PyPI on 26 August 2026, 43 days before the check (20 of 30). The deprecations page promises an email and a named replacement and states no minimum notice. The one speech-to-text retirement on it is `distil-whisper-large-v3-en` on 23 August 2025 (4 of 12). Neither Whisper model was retired or renamed in the last 90 days (8 of 8). A dated docs changelog with a feed exists, its newest entry is 18 April and none of its entries concern audio (8 of 15). The Python SDK repository had 3 open issues on 8 October. Replies were not sampled (5 of 10). Official `groq` 1.7.0 on PyPI and `groq-sdk` 1.6.0 on npm (15). CI, release automation and lock-file vulnerability fixes in 1.7.0 (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. Reliability, 90 out of 100, up to 2 more on the total

Why it scored 90: Hosted reading. The status page at groqstatus.com (incident.io) lists `whisper-large-v3` and `whisper-large-v3-turbo` as separate production components (20). On 8 October 2026 it showed both at 100% uptime for July to October 2026 and no open incident, and its feed held one item, a planned maintenance for another model on 9 October (30). The rate limits page gives numbers for both models, 20 requests a minute, 2,000 a day, 7,200 audio seconds an hour and 28,800 a day on the free plan, and the models page gives 300 requests a minute and 200,000 audio seconds an hour (v3) and 400 and 400,000 (Turbo) on the Developer plan (15). A 429 carries `retry-after`, the security onboarding page says to back off exponentially on 429 and 5xx, and transcription is a stateless call (15). The enterprise Performance Tier's 99.9% availability SLA names three language models and neither Whisper model, so no SLA was found for this product (0). Both models are listed as production models (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.

## 7. Transparency & trust, 85 out of 100, up to 1.3 more on the total

Made of editorial 72, provenance 98.

Why it scored 85: The hosted service is closed under the Groq Services Agreement. Both models are OpenAI's published Whisper weights, linked from each model card, served with Groq's own quantisation, and the SDKs are Apache-2.0 (20 of 30). The data page, the services agreement and the DPA agree on retention and on the training ban. The agreement limits zero retention to 'Eligible Customers' while the data page says all customers may enable it, and the privacy policy effective 12 November 2025 says it does not cover customer data sent to GroqCloud (26 of 30). A deprecations page with dated history and replacements, with no minimum notice period (14 of 20). Customer data is kept in Google Cloud storage in the United States, and the DPA gives 15 days' notice of sub-processor changes to subscribers. The sub-processor list is in the trust centre we could not read (12 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):

- Terms of service: read, states 7 of the 7 things a reader expects, and has 1 clause that costs points (8 of 10)

## 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: trust.groq.com is drawn by script, so certifications, any bug bounty and the sub-processor list were not read
- unchecked: groq.com/pricing is drawn by script. Prices come from the models page and the model cards, which agree
- unchecked: the status page's incident list before July 2026. Only the page as served and its linked feed were read, not the page's internal data feed
- unchecked: the groq.com registration date, taken from the 26 September check of the GroqCloud listing
- unchecked: replies on the SDK issue trackers
- The privacy policy says it does not apply to customer data processed through GroqCloud, which the services agreement and DPA govern. `provenance.privacy` points at it as on the GroqCloud listing, and a reader may prefer the DPA
- security.txt has a Contact line and no Expires field. It is recorded as valid, as on the GroqCloud listing
- The services agreement limits zero retention to 'Eligible Customers' while the data page says all customers may enable it
- The guide gives word error rates of 10.3% and 12%, and the Whisper Large v3 model card gives 8.4%, 10.0% and 11.0% by test type. None is our measurement
- The model cards list live captioning and realtime uses, and no streaming endpoint is documented. No deduction was taken because the guide itself makes no streaming claim
- The services agreement bars using the service to develop products that compete with it. No benchmarking clause was found
- The changelog's two newest entries carry no year. They follow 1 December 2025 and are read as 2026

## Weaknesses

- No streaming or realtime endpoint and no diarisation in the reviewed documentation
- Uploads are capped at 25 MB on the free plan and 100 MB on the Developer plan, so long recordings need client-side chunking
- `srt` and `vtt` response formats are not supported, and Whisper Large v3 Turbo cannot translate
- No OpenAPI document was found, and the docs changelog's newest entry is dated 18 April
- The 99.9% availability SLA of the enterprise Performance Tier names three language models and neither Whisper model

## 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 `whisper-large-v3-turbo` for transcription and `whisper-large-v3` for translation to English. The translations endpoint does not accept Turbo.
- Pass `url` instead of `file` for audio over 25 MB, and split anything over the plan's size limit into overlapping chunks before sending.
- Set `response_format` to `verbose_json` before asking for `timestamp_granularities[]`. Word timestamps add latency, segment timestamps do not.
- Every request is billed as at least 10 seconds of audio, so join very short clips where the task allows.
- Read `retry-after` on a 429 and back off. Audio limits count seconds an hour and a day as well as requests.

## 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.