What it does

Model Check is a free tool on AI Fast's developer documentation site for screening third-party relay services that offer an OpenAI-compatible model API. You enter the endpoint's base URL, a model ID and a temporary API key, choose a quick or standard run, and it sends a small set of fixed probe requests to that endpoint. The report shows what passed, looked doubtful or failed, item by item. Checks cover authentication and the model list, protocol compliance, instruction following, response metadata, token accounting, a randomly generated multi-step question, streaming output and tool calling. Writing style and the model's stated knowledge cutoff are recorded only as observations. A quick run takes about 3 requests and a standard run about 7. The site is explicit about limits. Results measure compatibility and sampled behaviour, cannot prove which model sits behind an endpoint, and are not a vendor certification. It says the API key is not stored in a database, cache or report, only public HTTPS addresses are accepted, and the test requests may incur small token charges on the endpoint being checked. Only OpenAI Chat Completions-compatible endpoints that use a Bearer key are supported, and runs are rate-limited.

Capabilities

Quick mode (about 3 requests) and standard mode (about 7 requests)
Itemised pass, doubtful or fail evidence rather than one opaque score
Checks for protocol, token fields, streaming and tool calling
Randomly generated multi-step questions, which the site says reduce the effect of tuning to fixed questions
Copyable report summary and README badge
Works with third-party endpoints, not only the vendor's own

When to use it

You are evaluating a third-party relay for an OpenAI-compatible model API
You suspect an endpoint is quietly swapping or downgrading the model you asked for
You want to confirm streaming and tool calling work before sending production traffic