GitHub Models to Shut Down Completely on July 30, Disabling Inference APIs and BYOK Endpoints

GitHub Models will be fully retired on July 30. The shutdown covers not only the web-based Playground, but also the model catalog, hosted inference APIs, and bring-your-own-key (BYOK) endpoints; existing customers still using the service will also be affected. GitHub has scheduled brief disruptions on July 16 and 23 so that workflows depending on these endpoints can expose failure paths early. After the final deadline, the relevant interfaces will be removed entirely.

This change is especially important for teams using GitHub Models as a lightweight model gateway, teaching environment, or CI testing backend. Even if an application uses an OpenAI-compatible request format, migrating to Microsoft Foundry may still involve new resource and deployment names, Azure identity permissions, regional availability, quotas, content filtering, and billing accounts. Identical model names do not guarantee that provider versions, default parameters, or response fields will match exactly. As a result, merely replacing the base URL could cause streaming parsers, tool calls, or structured outputs to fail at runtime.

Microsoft positions Foundry as an alternative model platform offering hosted and serverless inference, model comparison, and routing capabilities. However, GitHub has not promised automatic migration of existing configurations, keys, or usage records. Engineering teams should first inventory legacy endpoints in their code, GitHub Actions, sample projects, and secret stores. They should then replay real traffic to test latency, error codes, and output schemas before switching through dual writes or a rollback-capable configuration. Key issues to watch include exactly how the deadline will be enforced, whether failure responses will be consistent, and when SDKs, courses, and templates that still reference GitHub Models will be updated.