{
  "fixes": {
    "slug": "openai-speech-to-text",
    "name": "OpenAI Speech to Text",
    "listing": "https://www.anchorterminal.com/tools/openai-speech-to-text",
    "markdown": "# Fix list: OpenAI Speech to Text\n\nFrom Anchor Terminal's listing at https://www.anchorterminal.com/tools/openai-speech-to-text, the October 2026 research run, assessed 9 October 2026. Grade BB, 72.4 out of 100.\n\nThis 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.\n\nFor a coding agent working on OpenAI 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.\n\n## 1. Payments \u0026 pricing, 20 out of 100, up to 10 more on the total\n\nWhy it scored 20: No x402, MPP or L402 (0). $0.0045 an audio minute for `gpt-transcribe`, $0.017 for `gpt-live-transcribe` and $0.006 for `whisper-1` on the public pricing page (20). The rate limits guide names a Free tier with a $100 monthly usage limit, and the `gpt-transcribe` model page lists limits only from Build, which needs $5 of credit purchases. No free transcription allowance or no-card trial was found (0). A person signs up in a browser and creates the key in the console (0).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):\n\nThe published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).\n\n- 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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).\n\nPayment 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.\n\nOpen-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.\n\n## 2. Reliability, 80 out of 100, up to 4 more on the total\n\nWhy it scored 80: Hosted reading. status.openai.com (incident.io) lists Audio and Realtime as components of the API group (20). On 9 October 2026 it showed Audio at 100% uptime for July to October 2026. The linked feed holds five resolved incidents since 11 July that name the Audio component, all platform-wide elevated error rates (24 July, two on 25 July, 17 September and 6 October). Their durations were not read, and we score them as minor (20 of 30). The `gpt-transcribe` model page gives 5,000 requests a minute on Build, 10,000 on Launch and 30,000 on Grow (15). The rate limits guide documents `Retry-After` on 429 and 503, the `slow_down` and `server_is_overloaded` codes and backoff with jitter, and transcription is a stateless call (15). No SLA was found in the pages read. The Scale Tier page the guide links is on openai.com, which answered a bot check, so it is unread and scored absent (0). `gpt-transcribe` carries no preview label (10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):\n\nHosted APIs, MCP servers, models and platforms.\n\n- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 10, the surface agents use is generally available, not beta or preview.\n\nLocal packages, SDKs, frameworks and stdio MCP servers.\n\n- 20, installs from an official package with supported runtimes stated.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 15, version 1.0 or later, or declared stable.\n\nProtocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.\n\n## 3. Agent ergonomics, 78 out of 100, up to 3.6 more on the total\n\nWhy it scored 78: API reading, as for the other speech-to-text listings. The default `json` response is the text, detected languages and usage, `text` returns a bare string, and segments, words and speaker labels come only when asked for on the models that have them (20 of 25). There is no transcript store to page through. Files stop at 25 MB, the caller splits longer audio, and the `gpt-transcribe` model page marks the Batch API as not supported (10 of 20). Errors carry `message`, `type`, `param` and `code`, and the error codes page gives a fix for each status (18 of 20). The call is stateless and safe to retry, and the Python SDK retries 408, 409, 429 and 5xx twice by default. No idempotency key applies to this endpoint (15 of 20). `file` and `model` are the only required fields, with guide examples for JavaScript, Python, Go, Java, C# and Ruby SDKs and a CLI (15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):\n\n- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.\n\nModels 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.\n\n## 4. Transparency \u0026 trust, 61 out of 100, up to 3.4 more on the total\n\nMade of editorial 62, provenance 59.\n\nWhy it scored 61: The hosted models are closed. The Python SDK is Apache-2.0, the OpenAPI document is MIT, and the SDK says `whisper-1` runs the open-source Whisper V2 model. The service terms were not read, because openai.com answered a bot check (12 of 30). The data controls page states training, retention and application state for each endpoint. The privacy policy and any DPA were not read, so whether they agree with it was not checked (18 of 30). A deprecations page with notice periods by model class, dated shutdowns and replacements (20). Data residency is documented by region and endpoint, with the audio endpoints in all ten listed regions for storage and in the United States and Europe for processing. The sub-processor list the docs link answered a bot check (12 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):\n\n- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).\n\nThe other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.\n\nProvenance checks not met in full (half of this category, computed from checked facts):\n\n- Legal entity named: not found (0 of 20)\n- Domain age: openai.com, no registry record we could read (0 of 15)\n- Terms of service: published, but our reader couldn't read it (7 of 10)\n- Privacy policy: published, but our reader couldn't read it (7 of 10)\n\n## 5. Security \u0026 auth, 86 out of 100, up to 2.5 more on the total\n\nWhy it scored 86: Model reading. Bearer keys are created per project. Admin API keys are a separate credential that cannot call non-administration endpoints, a project can be limited to a list of models, and the error codes page documents an IP allowlist and keys without permission for an endpoint. The pages on service accounts and workload identity federation were not read (26 of 30). The data controls page says API data has not been used for training since 1 March 2023 unless the customer opts in (20). Both audio endpoints are listed with no abuse-monitoring retention, no application state and Zero Data Retention eligibility (15). The Admin API has an audit log endpoint for user actions and configuration changes (13 of 15). security.txt is PGP-signed, names a Bugcrowd programme and a coordinated disclosure policy and has no Expires field. Certifications were not read, because no page we read links a trust centre and openai.com answered a bot check (12 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-security):\n\n- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.\n\nModels 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.\n\n## 6. Maintenance \u0026 community, 75 out of 100, up to 2.2 more on the total\n\nWhy it scored 75: Model reading. The last dated change to transcription is the deprecation notice of 26 August 2026, 44 days before the check, and the release of `gpt-transcribe` and `gpt-live-transcribe` was on 28 July (20 of 30). The deprecations page promises at least six months' notice for generally available models, and the four transcription models got six months to the day (12 of 12). Four of the five file transcription models were deprecated within 90 days, a month after their replacements shipped, and the replacements do not cover diarisation, timestamps, subtitles or translation (2 of 8). The dated changelog has two transcription entries in 90 days among dozens for the API (13 of 15). The Python SDK repository showed 140 open issues and pull requests. Replies were not sampled (5 of 10). The Python SDK 3.26.1 was tagged on 8 October 2026, with 15 tags since 18 September (15). CI, CodeQL and a breaking-change check run in the repository (8 of 10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):\n\n- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.\n\nModels are read for deprecation notice periods and model churn rather than release counts.\n\n## 7. Schema \u0026 documentation, 88 out of 100, up to 2 more on the total\n\nWhy it scored 88: Model reading. A public OpenAPI 3.1 document in openai/openai-openapi covers `/audio/transcriptions` and `/audio/translations` and names `gpt-transcribe` (25). llms.txt and a Markdown twin of every docs page (10). The transcription overview says which model to use for files, live audio, speaker labels, timestamps and translation, and when not to use the specialised ones (18 of 20). `response_format`, `timestamp_granularities` and `model` are enums in the document. `language` is a plain string, and which formats each model accepts is stated in prose only (12 of 15). Examples in seven languages and curl, and an error codes page with a cause and fix for each. The Markdown twin of the endpoint reference omits the request parameters, its first example names the deprecated `gpt-4o-transcribe`, and the guide lists seven input formats where the document lists nine (11 of 15). A dated changelog and deprecations page. `gpt-transcribe` has one alias and no dated snapshot to pin (12 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):\n\nAPIs and MCP servers.\n\n- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 15, versioning and a public changelog.\n\nModels 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.\n\n## What we couldn't check\n\nWhat 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.\n\n- unchecked: the service terms, privacy policy, DPA and the contracting legal entity. openai.com answered the sub-processor page with a bot check (HTTP 403) on 9 October 2026 and no other openai.com page was requested after that. `provenance.terms` and `provenance.privacy` are left out\n- unchecked: the sub-processor list at openai.com/policies/sub-processor-list, which answered 403\n- unchecked: any SLA. The Scale Tier and Reserved Tier pages the rate limits guide links are on openai.com and were not requested\n- unchecked: certifications and a trust centre. No page read links one\n- unchecked: whether the Free tier can call `gpt-transcribe` and whether it needs a card. The model page lists limits from Build only\n- unchecked: the `gpt-live-transcribe`, `whisper-1` and `gpt-4o-transcribe-diarize` model pages, the service account and workload identity pages, and the Realtime sessions reference\n- unchecked: durations of the five status incidents that name Audio. Only the status page and its linked feed were read\n- unchecked: the openai.com registration date and npm and PyPI download counts\n- unchecked: replies on the SDK issue trackers\n- The deprecation of 26 August 2026 names `gpt-live-transcribe` or `gpt-transcribe` as replacements for `whisper-1` and `gpt-4o-transcribe-diarize`, and neither has diarisation, word timestamps, subtitle output or translation in the docs read. No deduction was taken because nothing has been removed yet and the notice is six months\n- security.txt has no Expires field. It is recorded as valid because it is signed and names two contacts\n- The lead named `gpt-4o-transcribe` and `whisper-1` among the product's models. Both are deprecated, and the listing name drops the model list\n- This product shares its API, key and status page with `openai-api`. If the owner holds listings whose governing terms were not read, this one qualifies\n- The rate limits guide's retry examples and the reference examples still name models past or near their shutdown dates\n- Robots.txt answers: developers.openai.com and openai.com 200 and allow the paths read, status.openai.com and api.github.com 404, read as no rules. developers.openai.com received 17 requests against the guide of about fifteen\n\n## Weaknesses\n\n- `whisper-1`, `gpt-4o-transcribe`, `gpt-4o-mini-transcribe` and `gpt-4o-transcribe-diarize` were deprecated on 26 August 2026 and shut down on 26 February 2027\n- The two named replacements return no speaker labels, word timestamps, `srt` or `vtt` output or English translation in the reviewed documentation\n- Uploads stop at 25 MB, the caller splits longer recordings, and the `gpt-transcribe` model page marks the Batch API as not supported\n- The Markdown twin of the endpoint reference lists the response fields and omits the request parameters, and its first example names a deprecated model\n- openai.com answered our reader with a bot check, so the service terms, privacy policy, sub-processor list and any SLA were not read\n\n## What costs an agent a turn today\n\nThe notes we give agents before they call it. Each one is a workaround an agent shouldn't need.\n\n- Send `gpt-transcribe` to `POST /v1/audio/transcriptions` for recorded files. Use `languages` (a list), not `language`, and never send both.\n- Keep each upload at 25 MB or less. Split longer audio between sentences and pass the previous chunk's text in `prompt`.\n- For speaker labels send `gpt-4o-transcribe-diarize` with `response_format=diarized_json` and `chunking_strategy=auto` for audio over 30 seconds. Plan for its shutdown on 26 February 2027.\n- Word timestamps, `srt`, `vtt` and `/v1/audio/translations` need `whisper-1`, which cannot stream and shuts down on the same date.\n- On 429 or 503 wait at least `Retry-After` when present, then back off with jitter. Do not retry `credit_balance_exhausted` or spend-limit errors.\n\n## When it's done\n\nSend 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.\n",
    "grade": "BB",
    "score": 72.4,
    "assessed": "2026-10-09",
    "run": "October 2026 research run",
    "categories": [
      {
        "key": "payments",
        "name": "Payments \u0026 pricing",
        "score": 20,
        "maxGain": 10,
        "reason": "No x402, MPP or L402 (0). $0.0045 an audio minute for `gpt-transcribe`, $0.017 for `gpt-live-transcribe` and $0.006 for `whisper-1` on the public pricing page (20). The rate limits guide names a Free tier with a $100 monthly usage limit, and the `gpt-transcribe` model page lists limits only from Build, which needs $5 of credit purchases. No free transcription allowance or no-card trial was found (0). A person signs up in a browser and creates the key in the console (0).",
        "checklist": [
          "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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-payments"
      },
      {
        "key": "reliability",
        "name": "Reliability",
        "score": 80,
        "maxGain": 4,
        "reason": "Hosted reading. status.openai.com (incident.io) lists Audio and Realtime as components of the API group (20). On 9 October 2026 it showed Audio at 100% uptime for July to October 2026. The linked feed holds five resolved incidents since 11 July that name the Audio component, all platform-wide elevated error rates (24 July, two on 25 July, 17 September and 6 October). Their durations were not read, and we score them as minor (20 of 30). The `gpt-transcribe` model page gives 5,000 requests a minute on Build, 10,000 on Launch and 30,000 on Grow (15). The rate limits guide documents `Retry-After` on 429 and 503, the `slow_down` and `server_is_overloaded` codes and backoff with jitter, and transcription is a stateless call (15). No SLA was found in the pages read. The Scale Tier page the guide links is on openai.com, which answered a bot check, so it is unread and scored absent (0). `gpt-transcribe` carries no preview label (10).",
        "checklist": [
          "Hosted APIs, MCP servers, models and platforms.",
          "- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 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.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-reliability"
      },
      {
        "key": "ergonomics",
        "name": "Agent ergonomics",
        "score": 78,
        "maxGain": 3.6,
        "reason": "API reading, as for the other speech-to-text listings. The default `json` response is the text, detected languages and usage, `text` returns a bare string, and segments, words and speaker labels come only when asked for on the models that have them (20 of 25). There is no transcript store to page through. Files stop at 25 MB, the caller splits longer audio, and the `gpt-transcribe` model page marks the Batch API as not supported (10 of 20). Errors carry `message`, `type`, `param` and `code`, and the error codes page gives a fix for each status (18 of 20). The call is stateless and safe to retry, and the Python SDK retries 408, 409, 429 and 5xx twice by default. No idempotency key applies to this endpoint (15 of 20). `file` and `model` are the only required fields, with guide examples for JavaScript, Python, Go, Java, C# and Ruby SDKs and a CLI (15).",
        "checklist": [
          "- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-ergonomics"
      },
      {
        "key": "transparency",
        "name": "Transparency \u0026 trust",
        "score": 61,
        "maxGain": 3.4,
        "reason": "The hosted models are closed. The Python SDK is Apache-2.0, the OpenAPI document is MIT, and the SDK says `whisper-1` runs the open-source Whisper V2 model. The service terms were not read, because openai.com answered a bot check (12 of 30). The data controls page states training, retention and application state for each endpoint. The privacy policy and any DPA were not read, so whether they agree with it was not checked (18 of 30). A deprecations page with notice periods by model class, dated shutdowns and replacements (20). Data residency is documented by region and endpoint, with the audio endpoints in all ten listed regions for storage and in the United States and Europe for processing. The sub-processor list the docs link answered a bot check (12 of 20).",
        "blend": "editorial 62, provenance 59",
        "checklist": [
          "- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-transparency"
      },
      {
        "key": "security",
        "name": "Security \u0026 auth",
        "score": 86,
        "maxGain": 2.5,
        "reason": "Model reading. Bearer keys are created per project. Admin API keys are a separate credential that cannot call non-administration endpoints, a project can be limited to a list of models, and the error codes page documents an IP allowlist and keys without permission for an endpoint. The pages on service accounts and workload identity federation were not read (26 of 30). The data controls page says API data has not been used for training since 1 March 2023 unless the customer opts in (20). Both audio endpoints are listed with no abuse-monitoring retention, no application state and Zero Data Retention eligibility (15). The Admin API has an audit log endpoint for user actions and configuration changes (13 of 15). security.txt is PGP-signed, names a Bugcrowd programme and a coordinated disclosure policy and has no Expires field. Certifications were not read, because no page we read links a trust centre and openai.com answered a bot check (12 of 20).",
        "checklist": [
          "- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-security"
      },
      {
        "key": "maintenance",
        "name": "Maintenance \u0026 community",
        "score": 75,
        "maxGain": 2.2,
        "reason": "Model reading. The last dated change to transcription is the deprecation notice of 26 August 2026, 44 days before the check, and the release of `gpt-transcribe` and `gpt-live-transcribe` was on 28 July (20 of 30). The deprecations page promises at least six months' notice for generally available models, and the four transcription models got six months to the day (12 of 12). Four of the five file transcription models were deprecated within 90 days, a month after their replacements shipped, and the replacements do not cover diarisation, timestamps, subtitles or translation (2 of 8). The dated changelog has two transcription entries in 90 days among dozens for the API (13 of 15). The Python SDK repository showed 140 open issues and pull requests. Replies were not sampled (5 of 10). The Python SDK 3.26.1 was tagged on 8 October 2026, with 15 tags since 18 September (15). CI, CodeQL and a breaking-change check run in the repository (8 of 10).",
        "checklist": [
          "- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.",
          "Models are read for deprecation notice periods and model churn rather than release counts."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-maintenance"
      },
      {
        "key": "schema",
        "name": "Schema \u0026 documentation",
        "score": 88,
        "maxGain": 2,
        "reason": "Model reading. A public OpenAPI 3.1 document in openai/openai-openapi covers `/audio/transcriptions` and `/audio/translations` and names `gpt-transcribe` (25). llms.txt and a Markdown twin of every docs page (10). The transcription overview says which model to use for files, live audio, speaker labels, timestamps and translation, and when not to use the specialised ones (18 of 20). `response_format`, `timestamp_granularities` and `model` are enums in the document. `language` is a plain string, and which formats each model accepts is stated in prose only (12 of 15). Examples in seven languages and curl, and an error codes page with a cause and fix for each. The Markdown twin of the endpoint reference omits the request parameters, its first example names the deprecated `gpt-4o-transcribe`, and the guide lists seven input formats where the document lists nine (11 of 15). A dated changelog and deprecations page. `gpt-transcribe` has one alias and no dated snapshot to pin (12 of 15).",
        "checklist": [
          "APIs and MCP servers.",
          "- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-schema"
      }
    ],
    "provenance": [
      {
        "label": "Legal entity named",
        "value": "not found",
        "points": 0,
        "max": 20
      },
      {
        "label": "Domain age",
        "value": "openai.com, no registry record we could read",
        "points": 0,
        "max": 15
      },
      {
        "label": "Terms of service",
        "value": "published, but our reader couldn't read it",
        "points": 7,
        "max": 10
      },
      {
        "label": "Privacy policy",
        "value": "published, but our reader couldn't read it",
        "points": 7,
        "max": 10
      }
    ],
    "unchecked": [
      "unchecked: the service terms, privacy policy, DPA and the contracting legal entity. openai.com answered the sub-processor page with a bot check (HTTP 403) on 9 October 2026 and no other openai.com page was requested after that. `provenance.terms` and `provenance.privacy` are left out",
      "unchecked: the sub-processor list at openai.com/policies/sub-processor-list, which answered 403",
      "unchecked: any SLA. The Scale Tier and Reserved Tier pages the rate limits guide links are on openai.com and were not requested",
      "unchecked: certifications and a trust centre. No page read links one",
      "unchecked: whether the Free tier can call `gpt-transcribe` and whether it needs a card. The model page lists limits from Build only",
      "unchecked: the `gpt-live-transcribe`, `whisper-1` and `gpt-4o-transcribe-diarize` model pages, the service account and workload identity pages, and the Realtime sessions reference",
      "unchecked: durations of the five status incidents that name Audio. Only the status page and its linked feed were read",
      "unchecked: the openai.com registration date and npm and PyPI download counts",
      "unchecked: replies on the SDK issue trackers",
      "The deprecation of 26 August 2026 names `gpt-live-transcribe` or `gpt-transcribe` as replacements for `whisper-1` and `gpt-4o-transcribe-diarize`, and neither has diarisation, word timestamps, subtitle output or translation in the docs read. No deduction was taken because nothing has been removed yet and the notice is six months",
      "security.txt has no Expires field. It is recorded as valid because it is signed and names two contacts",
      "The lead named `gpt-4o-transcribe` and `whisper-1` among the product's models. Both are deprecated, and the listing name drops the model list",
      "This product shares its API, key and status page with `openai-api`. If the owner holds listings whose governing terms were not read, this one qualifies",
      "The rate limits guide's retry examples and the reference examples still name models past or near their shutdown dates",
      "Robots.txt answers: developers.openai.com and openai.com 200 and allow the paths read, status.openai.com and api.github.com 404, read as no rules. developers.openai.com received 17 requests against the guide of about fifteen"
    ],
    "weaknesses": [
      "`whisper-1`, `gpt-4o-transcribe`, `gpt-4o-mini-transcribe` and `gpt-4o-transcribe-diarize` were deprecated on 26 August 2026 and shut down on 26 February 2027",
      "The two named replacements return no speaker labels, word timestamps, `srt` or `vtt` output or English translation in the reviewed documentation",
      "Uploads stop at 25 MB, the caller splits longer recordings, and the `gpt-transcribe` model page marks the Batch API as not supported",
      "The Markdown twin of the endpoint reference lists the response fields and omits the request parameters, and its first example names a deprecated model",
      "openai.com answered our reader with a bot check, so the service terms, privacy policy, sub-processor list and any SLA were not read"
    ],
    "agentNotes": [
      "Send `gpt-transcribe` to `POST /v1/audio/transcriptions` for recorded files. Use `languages` (a list), not `language`, and never send both.",
      "Keep each upload at 25 MB or less. Split longer audio between sentences and pass the previous chunk's text in `prompt`.",
      "For speaker labels send `gpt-4o-transcribe-diarize` with `response_format=diarized_json` and `chunking_strategy=auto` for audio over 30 seconds. Plan for its shutdown on 26 February 2027.",
      "Word timestamps, `srt`, `vtt` and `/v1/audio/translations` need `whisper-1`, which cannot stream and shuts down on the same date.",
      "On 429 or 503 wait at least `Retry-After` when present, then back off with jitter. Do not retry `credit_balance_exhausted` or spend-limit errors."
    ],
    "recheck": "https://www.anchorterminal.com/builders/#disputes"
  },
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-10-10",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.4",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run"
  }
}