DeepInfra

PASS consensus from 3 PASS.

The quickstart links API-key creation to complete cURL, Python, and JavaScript requests plus a sample response. One page carries the whole path, API key location, the Authorization: Bearer header, the base URL https://api.deepinfra.com/v1/openai, a named model, and working cURL, Python, and JavaScript calls with the expected response body. Authentication and the three-parameter path (base_url, api_key, model) are documented with ready-to-run curl, Python, and JavaScript examples.

PASS consensus from 2 PASS, 1 PARTIAL.

The limits page applies 200 concurrent requests per model to every account, with no separate plan quota listed. The concurrent default is exact and clearly scoped (200 per model, per account) and Batch API quotas are stated separately, but the Rate Limits page denies any per-minute limit while the account rate-limit endpoint returns a required tpm_rate_limit field described as "Per model token per minute rate limit" whose default value appears nowhere in the docs. The limit is stated as a single concrete number (200 concurrent per model), clearly scoped per-model, with an explanation of how multiple models aggregate.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The page gives delay, pacing, and increase actions, but it never defines the delay or a backoff schedule. All three recovery levers are named in one place, retry after a delay, slow the request rate, apply for an increase, but the delay is never quantified, no Retry-After header is documented, and the words "backoff", "exponential", and "jitter" appear nowhere in the docs, so any retry implementation is guesswork. The docs explain the 429 response (Rate limited), recommend retry with delay and pacing, and provide a dashboard path for requesting limit increases.

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

The page documents submission and success or failure payloads, but it gives no callback-authenticity verification method. Submitting the callback and handling both outcomes are fully documented with success and failure payloads and retry behavior, but authenticity verification has zero coverage, no signing secret, no signature header, no source IP range, and no mention of webhooks in DeepInfra's OpenAPI spec at https://api.deepinfra.com/openapi.json. Adding a webhook and receiving success/failure payloads is documented with code examples, but verifying callback authenticity (signing secret, signature header, HMAC scheme) is not documented anywhere.

PASS consensus from 3 PASS.

The official native page installs deepinfra and pairs its minimum TextGeneration example with an equivalent native cURL request. The "JavaScript client" section gives npm install deepinfra plus a runnable text-generation example whose constructor takes the same https://api.deepinfra.com/v1/inference/{model_name} endpoint and whose generate() arguments map one-to-one onto the native cURL body shown directly beneath it. The official deepinfra npm package is documented with a minimum TextGeneration example that maps directly to the native API's input/stop parameters.