GitHub Models が正式に廃止GitHub Models is now retired

GitHub has formally retired GitHub Models, the experimentation and prototyping environment it built for working with AI models. As of July 30, 2026, the service is no longer available to any customer, closing out a product that had allowed developers to browse, test, and integrate a range of models without leaving the GitHub ecosystem. For teams that had woven the offering into their workflows or early-stage projects, the shutdown is a hard cutoff rather than a gradual wind-down.

The retirement is comprehensive. According to GitHub's changelog, the four core components of the offering have all been switched off: the playground, the model catalog, the inference API, and the bring your own key (BYOK) capability. In practical terms, developers can no longer use the interactive playground to compare model responses, can no longer browse the catalog of available models, can no longer call models programmatically through the inference API, and can no longer supply their own provider credentials through BYOK. Because the removal applies to every customer, there is no tier or plan that retains access.

GitHub Models was positioned as a low-friction way to explore generative AI directly alongside code. The playground gave developers a browser-based space to prompt different models and observe how they behaved, which was useful for quick evaluation before committing to a particular model in production. The model catalog aggregated a selection of language models in one place, and the inference API let developers move from experimentation to code by calling those models from their own applications. The BYOK feature, added to broaden the service's flexibility, allowed users to plug in their own API keys rather than relying solely on GitHub-managed access, which helped with cost control and provider choice.

The move fits a broader pattern in which platform vendors consolidate overlapping AI capabilities. GitHub's flagship AI product remains GitHub Copilot, the coding assistant that has expanded well beyond autocomplete into chat, code review assistance, and agent-style features. GitHub is owned by Microsoft, whose Azure AI Foundry and related services provide model catalogs, hosted inference, and evaluation tooling at cloud scale. It appears likely that GitHub sees those Microsoft and Azure offerings, along with Copilot, as the more strategic homes for model access, though the changelog entry itself frames the change as a straightforward retirement rather than a rebranding.

For developers, the immediate consequence is that any integration depending on the GitHub Models inference API will stop functioning and needs to be re-pointed to an alternative. Teams that used the service for prototyping have several established options, including calling model providers such as OpenAI, Anthropic, Meta, Mistral, or Cohere directly, or routing through a hosted platform like Azure AI, Amazon Bedrock, or Google Vertex AI. The BYOK approach that GitHub Models supported translates fairly directly to those platforms, since most already center on developers bringing or provisioning their own credentials. Anyone who relied on the catalog purely for discovery can find comparable, and often larger, model listings on those hosted services.

It is worth distinguishing GitHub Models from the model-picking experience inside Copilot, which lets users select among different underlying models for chat and coding tasks. Those are separate surfaces, and the retirement announced here concerns the standalone GitHub Models product rather than Copilot's model options. The changelog does not indicate any change to Copilot as part of this retirement.

Retirements of this kind are a normal part of a platform's lifecycle, particularly for services that began as previews or experimental features aimed at gauging developer interest. The clear date and the explicit list of removed capabilities give teams a concrete reference point for auditing their own dependencies. Organizations that still have references to the playground, catalog, inference endpoints, or BYOK configuration in documentation, scripts, or continuous integration pipelines should verify that those paths have been migrated, since the endpoints are now expected to return errors rather than degrade gracefully. Developers seeking official guidance can consult The GitHub Blog and GitHub's documentation for any migration recommendations tied to the shutdown.