GitHub Copilot in VS Code gets BYOK: users can now connect their own keys for Anthropic, Gemini and OpenAI

GitHub announced that the Bring Your Own Key (BYOK) feature for language models is now generally available in the Visual Studio Code extension for GitHub Copilot. The feature is aimed at Copilot Business and Enterprise users and brings significant flexibility in model choice.

Which models can users now connect?

GitHub has greatly expanded the list of supported providers. Users can connect their own API keys from the following services: Anthropic (Claude models), Google Gemini, OpenAI, OpenRouter as an aggregator, and Azure AI for enterprise deployments.

In addition to cloud providers, GitHub has enabled support for local models. Supported options include Ollama for local execution of open-source models and Foundry Local, which is especially interesting for teams that cannot send code to external services for security reasons.

BYOK models are integrated into the complete Copilot experience — they work within the VS Code Copilot Chat and are available for use in custom agents that users configure for specific development tasks.

How does billing work with BYOK models?

The billing model is one of the key changes. When a user uses a BYOK model, costs go directly to the chosen provider — so Anthropic charges for its Claude models, OpenAI for its GPT, and so on.

Important: Copilot quota is not consumed when using a BYOK model. This means the user retains full Copilot license and its included models, with BYOK being an additional option rather than a replacement.

For enterprise teams this simplifies financial management because AI costs are not hidden within the Copilot subscription but appear as a separate line item with the provider.

Why does this matter for development teams?

Until now, teams wanting to use Claude for code within their editors were forced to use separate extensions like Claude Code or Cursor IDE, often running in parallel with Copilot. BYOK in VS Code consolidates that experience.

This is especially significant for organizations with existing Copilot agreements — they can now experiment with specialized models without changing their existing subscription. It is also relevant for teams that prefer certain models for certain tasks, for example Claude for reasoning and GPT for quick code generation.