Expanding Model Choice in VS Code with Bring Your Own Key (code.visualstudio.com)

VS Code has expanded model choice with a “bring your own key” (BYOK) workflow and a new Language Model Chat Provider API (v1.104) that lets third-party model providers ship models as VS Code extensions. BYOK (launched in March) already lets you plug in hundreds of models from providers like OpenRouter, Ollama, Google and OpenAI by supplying an API key. The new Provider API decentralizes that list so extensions can register models directly — available now to users on GitHub Copilot plans (Free, Pro, Pro+). Notable extensions include AI Toolkit (Azure AI Foundry & local Foundry models), Cerebras Inference (Qwen3 Coder and GPT OSS 120B at ~2,000 tokens/s, ~20x faster than many APIs), and the Hugging Face Provider (frontier LLMs like Kimi K2, DeepSeek V3.1, GLM 4.5). An OpenAI‑compatible provider is available in VS Code Insiders and can be tuned via the github.copilot.chat.customOAIModels setting. Why it matters: this turns VS Code into an extensible hub for local and cloud LLM experimentation, enabling fine‑grained control over which models and edit tools power chat and code workflows. It lowers friction for trying specialized or high‑performance inference providers, while retaining built‑in provider support. Caveats: BYOK doesn’t yet integrate with completions and model management UI is forthcoming. Planned improvements include better install flows, richer provider prompts, and a unified model management experience — signaling broader ecosystem growth but ongoing UX and integration work.