YuSMP Group logo
YuSMP Group Editorial Team Editorial team, YuSMP Group
Two glowing engines on a dark developer desk, one large and one small, sending streams of cyan light into a monitor of abstract code blocks, with a short stack of coins beside them

The short version

MAI Code 2 gives GitHub Copilot users a cheaper Microsoft-built option for multi-file coding, debugging and UI work, and Code 2 Flash covers quick edits and explanations. For teams running AI coding agents on usage-based Copilot billing, the model choice now shows up directly in the bill. Measure cost per finished task on your own code, not benchmark averages.

What did Microsoft announce at Ignite?

Microsoft AI, the division led by Mustafa Suleyman, introduced seven in-house MAI models covering code, cybersecurity, images, voice, transcription and decisions. Code 2 is the new flagship of its coding family, which started with the lighter MAI-Code-1-Flash earlier this year.

Microsoft splits the work between two models:

  • MAI Code 2: production-bound work such as multi-file implementation, deep debugging, repository-wide reasoning and UI.
  • MAI Code 2 Flash: fast, high-volume tasks such as code explanations, targeted edits, routine fixes and quick iterations.

Reuters framed the launch as Microsoft’s latest move to challenge Anthropic and OpenAI on cost. Suleyman told the agency that building models for specific domains lets Microsoft match rivals while keeping costs down. The Verge reported the same lineup, including the MAI Cyber 2 Flash security model.

How does MAI Code 2 compare with Claude and GPT?

Microsoft says MAI Code 2 comes close to the leading frontier coding models at a much lower price. Every figure below comes from Microsoft’s own testing.

ComparisonWhat Microsoft reports
MAI Code 2 vs GPT-6.1 Sol98% of the coding quality at about 40% of the price
MAI Code 2 vs Claude Opus 5.596% of the coding quality at 5% of the price
Blind frontend testDevelopers preferred MAI Code 2 over GPT-6 Sol 64% of the time
MAI Code 2 FlashComparable to GPT-6 Luna and Claude Haiku 5.5 on three routine coding benchmarks, at 65% of Luna’s cost and 20% of Haiku 5.5’s
Test setAverage of DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1

Read the price column with care. Microsoft says the Code 2 cost in its charts is a 50% promotional price, without giving an end date. When the promotion ends, the price gap with Claude Opus 5.5 narrows, from about 1:20 to about 1:10 if the list price is double.

Who can use MAI Code 2, and where?

Both models are generally available in GitHub Copilot from launch day. According to GitHub’s changelog, access depends on the plan:

  • MAI Code 2: Copilot Pro, Pro+, Max, Business and Enterprise.
  • MAI Code 2 Flash: the same plans plus Copilot Free and Student, on annual plans.

Developers can pick either model in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, the Copilot cloud agent, the Copilot app, github.com and GitHub Mobile. Copilot Fusion can also route tasks to them automatically. Microsoft says the models will reach OpenRouter, Vercel, OpenCode and Kilo later.

For Copilot Business and Enterprise, GitHub says new models are enabled by default unless an administrator turns off the global default or blocks specific models in the Copilot model policy.

What it means for US & EU software teams

Model choice is now a budget line. With Copilot on usage-based billing, routing routine agent steps to a model that costs a fraction of a frontier one can cut spend sharply. That only holds if the cheaper model finishes the task. A model that saves on tokens but needs a second attempt, or a human fix, costs more per merged change.

Default-on deserves a look. If your organization has rules on which AI vendors may see source code, a new model enabled by default can switch on before anyone reviews it.

Data terms need the same check. MAI models come from Microsoft rather than a third-party lab, which may suit teams already covered by Microsoft agreements. Still, confirm which GitHub and Microsoft data terms apply to MAI models on your plan before you allow them on repositories with client or regulated code. This matters most for EU teams under GDPR and for regulated industries.

Automatic routing also makes it harder to tell which model wrote a change. Teams that need an audit trail for AI-generated code should check what Copilot logs about the model used per session.

What should teams do this week?

  1. Review the model policy. In Copilot settings, decide whether MAI Code 2 and Code 2 Flash stay enabled by default for your organization.
  2. Run a short bake-off. Take 20 to 30 real tickets, run them on MAI Code 2 and on your current default, and record success rate, review comments and cost per merged change.
  3. Plan for the end of the promotion. Model your monthly Copilot spend at double the promotional price, so the bill does not surprise you later.
  4. Keep the guardrails. Tests, code review and least-privilege access for agents stay in place whichever model writes the code.

Frequently asked questions

What is MAI Code 2?

MAI Code 2 is Microsoft AI’s flagship coding model, announced at Microsoft Ignite on November 17, 2026. Microsoft built it for multi-file implementation, deep debugging, repository-wide reasoning and UI work. Its smaller sibling, MAI Code 2 Flash, handles code explanations, targeted edits and routine fixes.

How does MAI Code 2 compare with Claude Opus 5.5 and GPT-6.1 Sol?

Microsoft says MAI Code 2 reaches 96% of Claude Opus 5.5’s coding quality at 5% of the price, and 98% of GPT-6.1 Sol’s quality at about 40% of the price. The figures are averages over DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1, and Microsoft notes the Code 2 cost is shown at a 50% promotional price.

Which GitHub Copilot plans include MAI Code 2?

According to GitHub’s changelog, MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise, and MAI Code 2 Flash also on Copilot Free and Student. Both appear in the model picker and in Copilot Fusion routing, in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, github.com and GitHub Mobile.

Are new models like MAI Code 2 turned on automatically for Copilot Business and Enterprise?

Yes, unless an administrator changes the default. GitHub says new models are enabled by default unless admins disable the global default or block specific models in the model policy in Copilot settings. Teams with strict data or vendor rules should review that policy now.

Sources

Microsoft AI — MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices (primary source, November 17, 2026)
The Official Microsoft Blog — Ignite 2026: Building the frontier ecosystem (November 17, 2026)
GitHub Changelog — MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (November 17, 2026)
Reuters — Microsoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAI (November 17, 2026)
The Verge — Microsoft just announced a bunch of new AI models (November 17, 2026)