GitHub Copilot: GPT-5.3-Codex Now Generally Available

OpenAI's Latest Agentic Coding Model Arrives in Copilot

GPT-5.3-Codex, OpenAI's most advanced agentic coding model, is now generally available in GitHub Copilot as of February 9, 2026. The release marks a meaningful step forward in model capability for Copilot users across all subscription plans, with the model slated to become the default choice for many users as the gradual rollout completes.

The model's defining strengths are speed and reasoning in complex workflows. GitHub's early testing shows GPT-5.3-Codex delivers up to 25% faster performance than GPT-5.2-Codex on agentic coding tasks — a measurable improvement for developers using Copilot to run multi-step, tool-driven operations. The model also sets new benchmark highs across evaluations targeting coding capability, agentic reasoning, and real-world task performance.

Where Improved Reasoning Shows Up

The performance gains are most apparent in long-running, tool-driven workflows — exactly the scenarios where agentic AI earns its keep. When a coding agent must plan a sequence of file edits, run tests, interpret results, and iterate, the quality of its reasoning compounds across steps. GPT-5.3-Codex's improved execution in these scenarios makes multi-step tasks more reliable and reduces the need for developer intervention mid-run.

Availability Across the Platform

GPT-5.3-Codex is available to Copilot Pro, Pro+, Business, and Enterprise subscribers. The model integrates across the full range of Copilot entry points:

• Visual Studio Code (chat, ask, edit, and agent modes)
• GitHub.com
• GitHub Mobile (iOS and Android)
• GitHub CLI
• GitHub Copilot Coding Agent

For Business and Enterprise plans, administrators must enable GPT-5.3-Codex through the Copilot settings and policy configuration before team members can access it. The rollout is gradual, so availability may vary by account in the short term.