Fish Audio TTS API

# Fix list: Fish Audio TTS API

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/fish-audio-tts, the October 2026 research run, assessed 9 October 2026. Grade C, 60.7 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 Fish Audio TTS API: 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. Security & auth, 35 out of 100, up to 11.4 more on the total

Why it scored 35: Model reading of the checklist, with training and retention in place of the least-privilege and injection lines, as for the other TTS listings. Named, revocable API keys with optional expiry. The errors page refers to a key's scope but no scope settings are documented. The MCP server uses OAuth bound to one team. No short-lived token for browser TTS was found (20 of 30). The Terms of Use let Usage Data and Content train models with no opt-out found. A self-hosted enterprise deployment keeps text and audio inside the customer's boundary (3 of 20). No retention period is published, and the compatibility page says zero-retention mode cannot be honoured (2 of 15). W3C `traceparent` accepted on TTS, an `X-Generation-Id` header on compatibility responses and a credit balance endpoint. No per-call log documented (8 of 15). security.txt returned 404, the contributing page asks for security reports by email without giving an address, and no SOC 2, ISO 27001 or bug bounty was found in the pages read. The enterprise and trust pages on fish.audio were not read (2 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.

## 2. Payments & pricing, 30 out of 100, up to 8.8 more on the total

Why it scored 30: No x402, MPP or L402 (0). $15 per million UTF-8 bytes published without a login (20). `s2.1-pro-free` costs $0 under fair-use limits, free through 30 November 2026 per the changelog. Whether a card or a prepaid balance is needed is not stated (10 of 20). Signup is a browser flow with email verification and keys are created in the web app (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. Agent ergonomics, 67 out of 100, up to 5.4 more on the total

Why it scored 67: API reading. MP3, WAV, PCM or Opus with selectable sample rate and bitrate, chunked HTTP output, an SSE stream with alignment and two WebSocket endpoints (21 of 25). `GET /model` pages the voice library with `page_size`, `page_number`, title, tag, language and `self` filters and three sort orders. No maximum text length per request was found (14 of 20). Errors are JSON with `message` and `status`, a status table and typed SDK exceptions, with no error codes for TTS. An unrecognised `model` header falls back to `s2.1-pro` and an unresolved pronunciation dictionary is dropped, both without an error (11 of 20). Retry guidance for 429 and 5xx. No idempotency key, and the pricing page does not say whether a failed TTS request is billed, though it does for speech-to-text and voice design (9 of 20). Only `text` is required, and there are Python and JavaScript SDKs, though npm 0.1.0 defaults to the deprecated `s1` and Python defaults to `s2-pro` (12 of 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. Reliability, 75 out of 100, up to 5 more on the total

Why it scored 75: Hosted reading. Better Stack status page at status.fish.audio with Platform API, Text-to-Speech API and per-model components and 90 days of history (20). The Text-to-Speech API component shows downtime on 11 days in the last 90, the longest 58 minutes on 6 August and 40 minutes on 5 October 2026. Those follow the `s2.1-pro-free` component. Paid `s2.1-pro` shows 11 minutes on 15 July, and Platform API shows 2 hours 4 minutes on 12 July. One written incident, the APAC data centre failover on 10 August. No TTS outage reached an hour (18 of 30). Concurrency limits published as 5, 15 and 50 by prepaid amount (15). The docs say the native API returns 429 without `Retry-After` and tell callers to retry 429 and 5xx with exponential backoff. The compatibility endpoints send `Retry-After`. No idempotency key on TTS (12 of 15). The docs mention TTFA and DPA guarantees for `s2.1-pro` but no SLA document or figure was found (0). `s2.1-pro` is generally available, `drama-3-preview` is a preview (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.

## 5. Transparency & trust, 60 out of 100, up to 3.5 more on the total

Made of editorial 44, provenance 75.

Why it scored 60: Closed API with public terms, SDKs under Apache-2.0 and MIT, and S2-Pro weights published under Fish Audio's own research licence, which is not OSI-approved (20 of 30). Terms dated 18 August 2024 and a privacy policy dated 28 August 2024. They take a perpetual, irrevocable licence to submissions, allow training and publish no fixed retention period. The docs refer to DPA guarantees on `s2.1-pro`, and no DPA text was found (5 of 30). A deprecations page with dates, and S1's retirement announced 84 days ahead with a migration guide (16 of 20). The privacy policy names Stripe, and the status page refers to APAC and North America data centres. No sub-processor list found (3 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: fish.audio, registered 2023-12-11 (2 years) (7 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 3 clauses that cost points (3.1 of 10)
- security.txt: not found (0 of 10)

## 6. Maintenance & community, 69 out of 100, up to 2.7 more on the total

Why it scored 69: Read as a model. The changelog's newest entries are the S1 retirement notice of 8 October 2026 and `drama-3-preview` on 23 September 2026 (30). Three dated entries in the last 90 days, 18 September, 23 September and 8 October (20). Public changelog, Discord and a support address. We did not test how fast they answer (8 of 15). Official Python and JavaScript SDKs, last published on 10 March 2026. The Python fix for `s2.1` model names was merged on 31 July and is not in a tagged release, and npm 0.1.0 defaults to `s1` (7 of 15). The Python repository has CI across Python versions and Dependabot. Releases are 213 days old (4 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.

## 7. Schema & documentation, 84 out of 100, up to 2.6 more on the total

Why it scored 84: OpenAPI 3.1 file at docs.fish.audio/api-reference/openapi.json, and an AsyncAPI schema rendered on the WebSocket page. The standalone AsyncAPI link in llms.txt returned 404 (25). llms.txt, llms-full.txt, Markdown pages and two agent skills under `/.well-known/agent-skills/` (10). The guide says when to use `convert`, `stream` and the WebSocket, which model to pick and what each `latency` mode trades (16 of 20). Enums for `format`, bitrates, `latency` and the `model` header, ranges on `temperature`, `top_p` and `chunk_length`, one required field. `features` is a free list of strings and the `speed` range is in prose only (12 of 15). Examples in curl, Python and JavaScript and an errors page with a status table. The OpenAPI file lists only 401, 402 and 503 for `/v1/tts`, and the guide and the reference disagree on the defaults for `latency` and `chunk_length` (10 of 15). `/v1` paths and a dated changelog current to 8 October 2026, with no way to pin API behaviour (11 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.

## 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: the fish.audio pricing, enterprise and any trust or security pages. The Terms of Use forbid crawling or scraping the Services, so we read only the terms, the privacy policy and the docs host. A certification or DPA published there would raise the security and transparency scores.
- unchecked: GitHub star counts. The GitHub API refused our request for its rate limit, so `githubStars` is empty.
- unchecked: the PyPI project page, which answered with a short challenge page. The Python release date comes from the repository's tag and changelog.
- unchecked: `https://api.fish.audio/openapi.json`, which the docs call the canonical schema. We read the copy on the docs host and sent no request to the API host.
- Whether a failed or interrupted `/v1/tts` request is billed. The pricing page states the rule for speech-to-text and voice design only.
- Whether the free model needs a card or a prepaid balance, and what the fair-use limits are.
- What the TTFA and DPA guarantees on `s2.1-pro` consist of. No figure or document was found.
- The tool list of the MCP server. The MCP page points to one and shows none, so `toolCount` is empty.
- Whether a maximum text length applies to one `/v1/tts` request.

## Weaknesses

- The terms allow Usage Data and Content to train models, with no opt-out found
- An unrecognised `model` header falls back to paid `s2.1-pro` without an error
- The Text-to-Speech API status component shows downtime on 11 days in 90, the longest 58 minutes on 6 August 2026, mostly on the free model
- The published SDKs date from March 2026. `fish-audio` 0.1.0 on npm defaults to the deprecated `s1`
- No security.txt, certification, DPA text or sub-processor list found in the pages read

## 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 the `model` header on every request and check its spelling. A missing or unknown value is served and billed as `s2.1-pro`.
- Budget by UTF-8 bytes, not characters. Chinese, Japanese and Korean text costs about three bytes a character.
- Retry 429 and 5xx with exponential backoff. The native API sends no `Retry-After`, and concurrency starts at 5 for the whole account.
- Use `[bracket]` cues on the S2 family. `(parenthesis)` tags from `s1` are read aloud as text.
- Pass the model explicitly in the JavaScript SDK, because npm release 0.1.0 defaults to `s1`.

## 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.