Skip to content

GitHub Copilot adds MAI Code 2, which Microsoft says matches Claude Opus 5.5 at 5% of the cost

· by

The short version

GitHub added Microsoft's MAI Code 2 and MAI Code 2 Flash to Copilot's model picker on November 17, 2026, with Microsoft benchmarks claiming near-frontier coding performance at a fraction of GPT and Claude's cost.

GitHub Copilot adds MAI Code 2, which Microsoft says matches Claude Opus 5.5 at 5% of the cost

Microsoft put its own coding models into GitHub Copilot's model picker on November 17, 2026, and backed the launch with a specific number: MAI Code 2 scores 96% as well as Claude Opus 5.5 on a set of coding benchmarks while costing 5% as much, according to Microsoft's own test results.

What

GitHub added MAI Code 2 and the lighter MAI Code 2 Flash to the Copilot model picker the same day, per GitHub's changelog. MAI Code 2 targets multi-file edits, repo-wide reasoning, and debugging; Flash handles routine fixes and code explanations at lower cost. MAI Code 2 is available on Copilot Pro, Pro+, Max, Business, and Enterprise; Flash adds the Free and Student tiers, and both roll out only to subscribers on annual plans. According to Microsoft's own announcement, averaged across four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1), MAI Code 2 reaches 98% of GPT 6.1 Sol's score at 40% of the cost and 96% of Claude Opus 5.5's score at 5% of the cost. In blind side-by-side tests judged by 258 annotators, reviewers preferred MAI Code 2's frontend output over GPT-6 Sol 64% of the time. Microsoft says MAI Code 2 Flash matches GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their respective costs.

The discount pricing behind those numbers may not hold

Microsoft's own footnotes on those results note that the MAI Code 2 cost figures are shown "at its 50% promotional price," not a standing rate, so the gap to Claude Opus 5.5 and GPT 6.1 Sol could narrow once that promotion ends. Microsoft also says the models are "explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work," so the wins are scoped to everyday coding tasks, not frontier reasoning. For Copilot admins, both models are enabled by default under the existing model policy unless an organization has turned off automatic enablement, and Copilot bills them under its usage-based pricing, so a team on a fixed monthly request allotment will see any cost difference show up in its bill rather than in the model picker itself.

Context

The launch is part of a broader Microsoft Ignite 2026 wave: Microsoft shipped MAI Cyber 2 Flash and other new models the same day, continuing a push to put its own models directly into Copilot rather than offering it only as a wrapper around other labs' models. GitHub's changelog notes MAI Code 2 Flash is also in limited preview on Microsoft Foundry, with Open Router, Vercel, and other platforms planned.

What to watch next

Watch whether the benchmark edge survives once the 50% promotional price expires, and whether independent testing by developers outside Microsoft's own MSBench and Artificial Analysis evaluation harness confirms the gap once the models see wider day-to-day use.

Sources