GitHub Copilot adds model picker — users can now choose Claude, Codex, or Copilot as agent model

GitHub rolled out a model picker to all paid Copilot subscribers on February 26. Users can now select Claude, Codex, or the default Copilot model as the underlying engine for both inline completions and the autonomous coding agent. The agent mode spins up a GitHub Actions VM, clones the repository, makes changes, and opens a draft pull request. Users can assign the same GitHub issue to all three models simultaneously and compare the resulting PRs. Copilot reached 4.7 million paid subscribers as of January 2026, a 75% year-over-year increase.

Giving users model choice is an acknowledgment that no single model dominates all coding tasks. It also turns Copilot into a distribution layer for Anthropic and OpenAI rather than a pure competitor — GitHub benefits regardless of which model wins. For developers, the ability to run three agents on the same issue in parallel and compare outputs is a meaningful workflow change. Satya Nadella confirmed Copilot is now worth more to Microsoft than GitHub's entire $7.5B acquisition price — the model picker is designed to protect that position by avoiding lock-in as model preferences shift.