AI agents · OpenClaw · self-hosting · automation

Quick Answer

What Is MAI Code 2? Microsoft's Coding Model, Price, Access

Published:

The short answer

MAI Code 2 is Microsoft’s own coding model, launched November 17, 2026, priced far below frontier models and available today only inside GitHub Copilot (Flash also in a Foundry private preview). Microsoft claims it delivers 98% of GPT-6.1 Sol’s performance at 40% of the cost and 96% of Claude Opus 5.5’s performance at 5% of the cost, averaged over four coding benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1). These are vendor claims at promotional prices, not independent results.

MAI Code 2MAI Code 2 Flash
Built forDeep debugging, multi-file implementation, refactoring, repo-wide reasoning, UI and dashboardsExplanations, targeted edits, routine fixes, fast iteration
Copilot price per 1M tokens (in / cached / out), to May 17, 2027$0.50 / $0.025 / $2.50$0.05 / $0.005 / $0.25
Price from May 17, 2027$1.00 / $0.05 / $5.00$0.10 / $0.01 / $0.50
Copilot plansPro, Pro+, Max, Business, EnterpriseFree, Student, Pro, Pro+, Max, Business, Enterprise
Outside CopilotNot yetMicrosoft Foundry private preview
Microsoft’s claim96% of Opus 5.5 at 5% of costComparable to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of cost

For context on the other models in the Copilot picker, see our current API prices table.

What Microsoft says it is good at

Microsoft positions MAI Code 2 for “the coding work our customers do every day” and says outright that the models are not yet tuned for harder problems such as fundamental science research and very long-horizon agentic work. It also highlights frontend work: in blind evaluations of generated frontend apps, reviewers preferred MAI Code 2’s output over GPT-6 Sol’s 64% of the time. Launch partners include Woven by Toyota, which reports that a MISRA-tuned MAI Code 2 Flash produced merge-ready fixes for 44% of cases where its frontier models needed human rework, at under a tenth of the cost.

How it compares on price

At its promotional rate, MAI Code 2 costs one-eighth of Claude Opus 5.5 in Copilot ($4 / $20 per million tokens) and a quarter of GPT-6.1 Sol and Claude Sonnet 5.5 (both $2 / $10). Even at the full price from May 2027 ($1 / $5) it matches GPT-6.1 Terra. MAI Code 2 Flash, at $0.05 / $0.25, is half the price of Claude Haiku 5.5 and GPT-6 Luna ($0.10 / $0.50 each).

For a Copilot user this means credits last much longer. A typical agent task with 200,000 input and 20,000 output tokens costs about 15 AI credits on MAI Code 2 against about 120 on Opus 5.5. Our Copilot plan comparison shows what that does to each plan’s allowance.

Should you switch to it?

  • Yes, for routine work in Copilot: bug fixes, small features, refactors, tests and UI tweaks. Try it as the default and see where it fails.
  • Keep a frontier model for hard problems: long agent runs, subtle concurrency bugs or unfamiliar codebases; Microsoft itself says MAI Code 2 is not built for these yet.
  • Admins: Copilot Business and Enterprise enable new models automatically under default model enablement; turn it off in the Copilot model policy if you need to review first.
  • Building your own agent: wait for broader Foundry availability; for now, compare options in best AI coding models ranked by benchmark and cost.

Microsoft announced seven MAI models at Ignite on November 17, 2026, including MAI Cyber 2 Flash for verified security defenders; MAI Code 2 is the one most developers will meet first, because it is already in their editor.

Last verified: November 18, 2026. Prices from GitHub’s Copilot models-and-pricing table; benchmark claims are Microsoft’s own.

Sources