Claude Check

An OpenAI-compatible endpoint that says it is Claude is not always Claude. This probe set is specifically tuned to discriminate Anthropic-trained Claude from non-Claude substitutes, and to estimate the size class (Opus / Sonnet / Haiku) and version line (3.5 / 3.7 / 4.5 / 4.6 / 4.7).

Scope and honest limits

Vendor verdict is the most reliable layer; size is medium reliability; version is the weakest — gateways routinely route a tier without exposing the point version, and Anthropic's own self-identification has drifted across releases. Treat the version verdict as a hint, not a ruling.

This probe path uses the proxy route only. The Anthropic native API (/v1/messages) is not supported — bring an OpenAI-compatible gateway URL.

Endpoint configuration

Without the trailing /chat/completions. Example: https://api.openrouter.ai/v1

Gateway base URL

API key

Sent only to /api/proxy; never logged or persisted.

Claimed model id

What the gateway calls the model. Example: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5.

No results yet. Configure the endpoint above and run.

Heuristic. Results are signal aggregation, not certified attestation. Cross-check with the main auditor and your own observability.