GitHub will fully retire GitHub Models on July 30, 2026. After that date, the playground, model catalog, inference API, and bring-your-own-key endpoints will no longer be available to any customer, including existing users with active usage.
The company also set two brownout dates before the shutdown. GitHub Models requests will temporarily return errors on July 16 and July 23, then service will be restored until the final cutoff.
That turns GitHub Models from a soft product retreat into a short migration window. In June, GitHub closed the service to new customers. The July 1 update gives existing customers the deadline.
The affected surface is wider than the playground
The playground is the visible part of GitHub Models, but it is not the only part teams need to inventory. GitHub’s retirement notice names the model catalog, inference API, BYOK endpoints, and related UI.
That matters because developers may have used GitHub Models in different ways. Some used it to compare models inside GitHub. Others may have wired prototype applications or internal tools to the inference API. Teams that brought their own provider keys now need to check whether any local workflow depends on GitHub as the routing layer.
The brownouts are the practical test. If requests fail during the July 16 or July 23 interruption, the team has found a dependency that needs a replacement before July 30.
GitHub points users to Azure AI Foundry and Copilot
GitHub’s migration guidance separates two jobs. For projects that need AI model access, it points to Azure AI Foundry. For AI-powered workflows directly on GitHub, it points to GitHub Copilot.
Those are not identical replacements. Azure AI Foundry is the more natural home for model catalog, endpoint, and production workflow needs. Copilot is a developer-assistance product with model access inside coding surfaces.
The distinction is important for teams planning the migration. If GitHub Models was being used as an experimentation tool, Azure AI Foundry may be the next platform. If it was being used as a lightweight model endpoint, the move may require billing, identity, endpoint, and governance work that was not part of the original GitHub setup.
The deadline is also a platform signal
GitHub has been expanding Copilot’s model picker, enterprise model policies, auto model selection, and managed settings. At the same time, GitHub Models is disappearing as a separate model playground and API surface.
The pattern is clear: GitHub is concentrating model choice inside Copilot for developer workflows, while broader model building and deployment moves toward Azure AI Foundry.
That is a sensible product boundary for Microsoft, but it changes the user promise. A convenient model playground inside GitHub is becoming a cloud-platform migration.
