VSCode Blogs

Microsoft's VS Code team published findings from 50,000 runs of a 5-line evaluation, revealing how AI coding models calibrate effort, token cost, and tool use. The blog also announced new features including bring-your-own-key support for language models and token efficiency improvements for GitHub Copilot.

From the VS Code blog /blogs/2026/06/19/what-50000-runs-taught-us Latest What 50,000 Runs of a 5-Line Eval Taught Us How AI coding models calibrate effort, token cost, and tool use on even the simplest task, and what that means for model selection and cost. Read post → Recent posts /blogs/2026/06/18/byok-vscode Use your own language model key in VS Code Learn how to use bring your own key BYOK in VS Code to add models from providers like Azure, Anthropic, Gemini, OpenAI, Huggingface, OpenRouter, or use a local model with Ollama, Foundry Local, and more. /blogs/2026/06/17/improving-token-efficiency-in-github-copilot Improving token efficiency for GitHub Copilot in VS Code Learn how we're improving token efficiency for GitHub Copilot to reduce costs and latency for VS Code users. /blogs/2026/05/15/agent-harnesses-github-copilot-vscode The Coding Harness Behind GitHub Copilot in VS Code Learn why the coding harness around GitHub Copilot in VS Code matters as models, tools, agents, and providers evolve.