GitHub Models Retired on July 30: Where to Go Next

GitHub shut down its Models service on July 30, 2026, removing access to the playground, model catalog, inference API, and bring-your-own-key (BYOK) feature for all customers, including those with active usage. The closure completes a retirement process GitHub announced on July 1. Developers who relied on GitHub Models for direct AI model access will need to move to Microsoft Foundry or GitHub Copilot depending on their use case.

What happened

Detail	Fact
Retirement date	July 30, 2026
Retirement announcement date	July 1, 2026
Features removed	Playground, model catalog, inference API, BYOK
Customers affected	All customers, including those with active usage
Replacement for model access	Microsoft Foundry
Replacement for GitHub-native AI workflows	GitHub Copilot

GitHub pulled the plug on GitHub Models on July 30, 2026. Every part of the service is gone: the browser-based playground for testing models, the model catalog, the inference API that developers called from their own code, and the BYOK option that let teams use their own API keys. No customer is exempt, according to GitHub’s changelog.

The move was not a surprise. GitHub flagged the shutdown on July 1, giving teams roughly four weeks to migrate. Whether that window was enough depends on how deeply a project was integrated with the inference API.

What are the alternatives to GitHub Models?

GitHub points to two paths forward, and they serve different needs:

• Microsoft Foundry: Described by GitHub as offering a “broad model catalog.” This is the closest like-for-like replacement if you want to browse and call hosted AI models via an API.
• GitHub Copilot: If your workflow lives inside GitHub (pull request reviews, code suggestions, repository automation), Copilot still provides access to a range of models. It is not a direct API replacement for arbitrary inference calls.

Teams building custom AI integrations, such as retrieval-augmented generation pipelines or automated content tools, will most likely need to move to Microsoft Foundry or another inference provider. If you’re looking at broader AI integration options for your stack, the choice of inference provider matters more than it might seem at first.

Why it matters

GitHub Models was a convenient on-ramp. It gave developers a no-friction way to test models and wire them into GitHub Actions or other workflows without setting up a separate account elsewhere. Losing that means one more platform to manage and one more billing relationship to maintain.

The consolidation also signals Microsoft’s broader strategy: route AI model consumption through Foundry, and keep Copilot focused on the developer experience inside GitHub. The two products are being kept separate rather than merged.

For teams that were using GitHub Models in a lightweight or experimental capacity, the migration is probably straightforward. For anyone running production inference through the API, there may be credential changes, endpoint changes, and quota differences to work through on Microsoft Foundry.

Our take

Four weeks is a tight notice window for a production service. GitHub Models was marketed as a way to build AI-powered apps directly on the platform, so teams that took that seriously are now scrambling. The fact that even customers with active usage got no grace period past July 30 is blunt.

Microsoft Foundry is a capable replacement, but it is a different product with its own setup curve. If you were using BYOK, check whether Foundry’s key management works the same way for your security requirements before assuming it’s a drop-in swap.

We have seen this pattern before with developer preview services: a convenient free or low-cost API gets pulled when the parent company decides to consolidate around a flagship product. If your business depends on a hosted inference API, always have a fallback provider configured. The GitHub Copilot angle is fine for IDE and code review use, but it does not replace general-purpose model inference for workflow automation or custom app backends.

If you are still figuring out which AI service fits your project, reviewing what you actually need from a model API before migrating will save time later.

Source: GitHub Changelog