Microsoft launches MAI Code 2 and MAI Code 2 Flash in GitHub Copilot
Microsoft has released MAI Code 2 and MAI Code 2 Flash, two coding models built by its own Microsoft AI (MAI) team, announced at Ignite on November 17, 2026. Both are available in GitHub Copilot now, from VS Code to the Copilot CLI. For developers this matters because Microsoft is pitching them on cost, not on topping the leaderboards: close to the frontier models from OpenAI and Anthropic, for a fraction of the price.
What’s new
- Two models. MAI Code 2 is the general-purpose model for harder work: production code, deep debugging, multi-file changes, refactoring and reasoning across a whole repository, plus UI work. MAI Code 2 Flash is the lighter, cheaper one for high-volume tasks: explaining code, targeted edits, routine fixes and quick iterations.
- What they’re not for. Microsoft says neither model is tuned for fundamental science research or very long-horizon agentic work.
- Benchmarks. Averaged over DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1, Microsoft says MAI Code 2 reaches 98% of GPT 6.1 Sol’s score at 40% of its cost, and 96% of Claude Opus 5.5’s score at 5% of its cost. In blind human evaluations with 258 annotators, covering UIs, games, data visualizations, 3D scenes and SVG, reviewers preferred MAI Code 2 over GPT-6 Sol 64% of the time. For Flash, Microsoft claims “comparable performance to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of the cost, across three coding benchmarks”.
- How they were measured. Microsoft says the runs used maximum reasoning effort. Three of the benchmarks ran on its MSBench harness with the GitHub Copilot CLI and no subagents; FrontierCode scores and costs come from Artificial Analysis. These are Microsoft’s own numbers, not independent results.
- Pricing. The cost figures in Microsoft’s charts use a 50% promotional price for both models. Microsoft’s post doesn’t give list prices.
- Also inside Microsoft 365. A further-tuned version of Flash, MAI Work Flash, serves Microsoft 365 and Fabric IQ. In Word, Microsoft says it performs close to GPT 6.1 Sol on five internal benchmarks while sending about 42% fewer input tokens per call in production.
What it means for developers
In GitHub Copilot, eligible subscribers can pick them now. GitHub’s changelog says: “Both models are available to eligible GitHub Copilot subscribers on annual plans.” According to the same changelog, MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise. MAI Code 2 Flash also reaches Copilot Free and Student, so it is the one free-tier users can try. Both show up in the model picker in:
- Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse
- Copilot CLI, the GitHub Copilot app and the Copilot cloud agent
- github.com and GitHub Mobile on iOS and Android
GitHub says both models are billed under usage-based billing; the per-model rates are on its models and pricing docs, where both are listed as generally available. Business and Enterprise admins control access through the model policy in Copilot settings. Under default model enablement, new models turn on automatically unless an admin has switched off the global default or disabled the model, so check that policy if your organization limits which models code can be sent to.
Outside Copilot, access is narrower. MAI Code 2 Flash is in a limited-access preview in Microsoft Foundry. Microsoft says it will release the model on OpenRouter, Vercel, Hermes, Open Code and Kilo soon, without dates. If you build on Foundry, Microsoft also points to its custom models page for customizing them.
Try it on your own tasks. The headline numbers are averages from Microsoft’s runs. Since both models sit in the same Copilot picker as the GPT and Claude models, the quickest check is to run your usual tasks (a failing test to fix, a multi-file refactor) with MAI Code 2 and the model you use now, and compare the results and the usage bill.
How it compares
Microsoft compares MAI Code 2 with GPT 6.1 Sol and Claude Opus 5.5, and Flash with GPT-6 Luna and Claude Haiku 5.5. OpenAI’s newer GPT-6.2 Astra and Sol, released the same day, are not in Microsoft’s comparison. Copilot also retired several older models this week; our post on the Copilot models deprecated on November 16 lists them.
Sources
- Introducing MAI Code 2 and MAI Code 2 Flash (Microsoft AI)
- MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (GitHub changelog)
- Supported AI models in GitHub Copilot (GitHub Docs)