We’re excited to announce that you can now bring your own language model into Visual Studio Chat.

You can connect API keys from providers (OpenAI, Anthropic, and Google) to access a wider range of models. This makes it easy to test the latest releases, customize workflows, or run on infrastructure you control.

What it unlocks

• Choice – Access more than just the built-in defaults, and try new models the moment they drop.
• Customization – Pick models that align with your security, infra, or performance needs.
• Control – Manage usage, quotas, and billing directly with your provider.
• Flexibility – Switch seamlessly between Copilot’s built-in models and your own.

Getting started

If you already have an API key from OpenAI, Anthropic, or Google, setup takes just a minute:

1. Open the Chat Window in Visual Studio.
2. Select Manage Models from the model picker.
3. Choose your provider and paste your API key.
4. Pick from the preset list, or enter a model name.
5. That’s it—your model will now show up in the picker.

A few important considerations:

• This feature currently applies only to Chat in Visual Studio (not completions, commit messages, or other AI features).
• Model capabilities vary—some may not support advanced features like tool use or vision inputs.
• Certain services (embeddings, repo indexing, intent detection) will still run through Copilot’s API.
• Output from third-party models comes directly from the provider and may not pass through Copilot’s responsible AI filters.

This is just the beginning. We’re expanding support for additional model providers soon, and we look forward to seeing how you use this feature and which models you choose.