MAI Code 2

MAI Code 2 is Microsoft AI’s second-generation coding model, announced at Microsoft Ignite on 2026-11-17 as the main model of the MAI coding line. It follows MAI-Code-1-Flash, which Microsoft offered in GitHub Copilot from Build 2026. It is the larger sibling of mai-code-2-flash. Part of the MAI series.

Positioning (Microsoft’s post)

  • Aimed at enterprise coding: taking code to production, deep debugging, multi-file implementation, reasoning across a whole repository, and UI work.
  • Microsoft states it is not tuned for fundamental science research or for very long-horizon agentic work.

Microsoft’s claims

Benchmarks named in the post are DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode 1.1. All models ran at maximum reasoning effort; DeepSWE, Terminal Bench and SWE-Atlas ran on MSBench with the GitHub Copilot CLI, without subagents. FrontierCode figures and costs came from Artificial Analysis.

  • Against GPT-6.1 Sol: 98% of the performance at 40% of the relative cost.
  • Against Claude Opus 5.5: 96% of the performance at 5% of the relative cost.
  • Frontend work: in blind app evaluations with 258 annotators, developers preferred MAI Code 2 over GPT-6 Sol 64% of the time, across UIs, games, data visualisations, 3D scenes and SVG graphics.

Availability

  • GitHub Copilot: available.
  • Other channels: OpenRouter, Vercel, Hermes, OpenCode and Kilo are listed as coming soon, with no dates.
  • Microsoft Foundry: the post names only mai-code-2-flash as available there, in limited-access preview, so this note does not claim Foundry access for Code 2.
  • A page on customising MAI Code 2 is linked from the post.

Billing

The post gives costs only as relative percentages. Its footnotes mention a 50% promotional rate for both Code 2 models and do not give absolute prices or an end date. Check Microsoft’s pricing page for current billing names and tiers.

Source limits

  • The post states no context window, parameter count or model card, and it does not state general availability for an API.
  • All performance and cost figures are Microsoft’s own, under the conditions above.

Sources