[
](/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.
source & further reading
code.visualstudio.com — original article
Visual Studio Code 1.126
What 50,000 Runs of a 5-Line Eval Taught Us
Use your own language model key in VS Code