Visual Studio Code expands its Artificial Intelligence capabilities with flexible language model selection for both chat and code completions. Users can switch models directly from the chat input’s model picker, where availability varies by chat mode and organizational settings. In agent mode, only models that support tool calling appear. For organizations on Copilot Business or Enterprise, administrators must enable certain models via Editor Preview Features. Paid Copilot users see premium request multipliers for eligible models, and the model list shown in the picker can be customized via Manage Models.
A new preview feature, auto model selection, is rolling out with VS Code 1.104. When enabled, VS Code dynamically chooses a model to maintain performance and reduce rate limiting, detecting degraded performance and switching to a better option. Auto currently selects among models such as Claude Sonnet 4, GPT-5, GPT-5 mini and others, while respecting any organizational opt outs. If premium access is exhausted, auto falls back to a 0x multiplier model. Users can hover over a chat response to see the active model and multiplier. Code completions use a separate model setting that can be changed from the Chat menu, and the set of completion models will evolve over time.
Beyond the built-in options, VS Code supports bring your own language model key for chat, giving access to a wide range of providers and models, including local deployments. Benefits include broader model choice, faster experimentation, local compute, and more control over provider limits. This feature is not yet available for Copilot Business or Enterprise. Providers can be configured via Manage Models by adding API keys, endpoints, and specific model selections. For agent mode, models must support tool calling to appear. VS Code also supports a custom OpenAI-compatible provider in VS Code Insiders 1.104, allowing users to define endpoints, capabilities such as vision and tool calling, and token limits, or configure models via the github.copilot.chat.customOAIModels setting.
There are important considerations. Bring your own key applies to chat only and does not affect code completions or other Artificial Intelligence features such as commit message generation. The Copilot API remains in use for tasks like embeddings, repository indexing, and intent detection. Responsible Artificial Intelligence filtering is not guaranteed when using external models. Locally hosted models can be used in chat through supported providers, extensions like AI Toolkit, or an OpenAI-compatible setup, but they are not available for code completions. Currently, using local models still requires a Copilot plan and an internet connection.
