ElevenLabs

PASS consensus from 3 PASS.

The quickstart covers key creation, installation, environment loading, conversion, playback, and execution in one sequence. The quickstart gives one linear path with a direct dashboard link for the key, pip install elevenlabs, a complete runnable Python snippet with a working voice ID, and a run command; saving to an .mp3 file is covered in the streaming guide linked from Next steps. The quickstart provides a four-step path (create API key, install SDK, write code, run) with a complete working example including import statements and the play() call.

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

The 429 page gives unscoped plan concurrency values, while Models adds doubled Flash limits; two alternative concurrency URLs failed. No page in the documentation publishes a per-plan monthly credit quota (the billing page defers to the off-docs pricing page, and the only quota figure anywhere is an incidental "30,000" in the rollover article), while the Text to Speech concurrency values and the per-request character limits each appear on two pages that disagree; the sentence quoted above also says five plans and then names six. Monthly credit quotas (10K/30K/121K/600K/1.8M/6M) are only on the interactive pricing comparison table, not in a docs page. Concurrency and per-request limits are well documented. Rollover rules contradict: the main docs page says all credits roll over up to two months, while the help center page says credit rollover does not apply to accounts on the free tier.

PARTIAL consensus from 3 PARTIAL.

The general guide gives both corrections, but its 429 labels differ from the dedicated 429 page. The errors reference cleanly separates the two 429 causes and prescribes exponential backoff for rate limiting and waiting for in-flight requests for concurrency, but the help-center 429 article names the concurrency error too_many_concurrent_requests instead of concurrent_limit_exceeded, so an agent branching on the error string picks the wrong constant depending on which page it read. The three 429 error types (rate_limit_exceeded, concurrent_limit_exceeded, system_busy) are clearly distinguished with type classifications and actionable guidance. However, no documented retry behavior exists, no Retry-After header, no exponential backoff recommendation, no SDK example for handling rate limit errors.

PASS consensus from 3 PASS.

The guide documents HMAC creation and verification; the API HTML URL returned 401, while its Markdown mirror loaded. The create endpoint takes auth_type "hmac" and returns webhook_secret, and the FastAPI example reads the raw body, passes it with the elevenlabs-signature header to construct_event, returns 401 on failure and 200 on success; the raw signature encoding is not documented, so verification without an official SDK is not possible from these pages. The webhook verification process is fully documented: HMAC signature setup, SDK construct_event method (Python and JS), required header verification, raw body requirement, HTTP 200 response, and idempotency guidance. Complete FastAPI example provided.

PASS consensus from 3 PASS.

The Python example installs the SDK, sets the key, and sends REST-aligned text, voice, model, and format fields. The Python SDK arguments map one to one onto the documented REST path parameter, query parameter, and body fields on the same page, and the client construction matches the authentication page, though the import is written from elevenlabs import ElevenLabs here and from elevenlabs.client import ElevenLabs in the quickstart. The SDK installation command, client initialization with API key, and text-to-speech convert method all match the REST API reference (convert endpoint). The example uses the same parameters (text, voice_id, model_id, output_format) documented in the API reference.