What is GitHub Copilot’s "Raptor mini" and why should devs care?
SMRTR summary
Developers armed with curiosity and debug logs have uncovered what GitHub's mysterious new "Raptor mini" actually is, after the company quietly slipped it into Copilot with virtually no explanation.
The model turns out to be a fine-tuned GPT-5-mini, customized by Microsoft specifically for coding tasks. What makes it surprising? For something called "mini," it packs a hefty 264,000-token context window and can output 64,000 tokens at once.
By digging through VS Code debug logs, developers discovered Raptor mini is designed for workspace-scale tasks like applying changes across multiple files or handling lengthy error messages. It's served through GitHub's Azure infrastructure and built to work seamlessly with Copilot's tools and agents.
The model appears to be GitHub's way of testing a code-focused GPT-5 variant in the wild. It's rolling out gradually to Copilot Pro users in VS Code, where it excels at practical developer tasks rather than creative writing.
Early testing shows it handles tool calls effectively and maintains decent speed at around 122 tokens per second, making it ideal for real coding workflows.
SMRTR provides this summary for quick context. The original article belongs to Dev.to.
Read the original article