Observed arrival · 2026-10-06
ModelsExam checks whether model APIs behave as advertised
An open-source API checker sends fixed tests to Claude, OpenAI-compatible, or image-model endpoints and reports on behavior, protocol conformance, and usage claims.
- For
- Operators checking third-party model API endpoints
- Worth noticing
- A run can make more than twenty upstream requests, take up to ten minutes, and incur provider charges.
Field notes
The workflow starts by selecting a protocol and choosing a model from the endpoint’s /v1/models listing, then compares fixed test responses with named official baselines. The page says a full run can take up to ten minutes and involve more than twenty upstream requests, which may create provider charges. Its badges show only a recent conclusion and expire after 30 days; the site says reports are not publicly listed unless their links are shared.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue