Skip to content

MAI Code 2: Microsoft's cheap coding model lands in GitHub Copilot

Comic-style small white robot coding on a holographic keyboard beside an idle gold robot on a throne of coins Models

Microsoft MAI Code 2 is a new coding model from Microsoft AI, released on November 17, 2026, at the Ignite conference. It arrived together with MAI Code 2 Flash, a smaller and cheaper version for routine work. Both models are already available in GitHub Copilot.

Microsoft's pitch is price. It says MAI Code 2 delivers 98% of the coding performance of OpenAI's GPT-6.1 Sol at about 40% of the cost. Against Anthropic's Claude Opus 5.5, it claims 96% of the performance at 5% of the cost.

What MAI Code 2 and Code 2 Flash are for

Microsoft built both models for the coding that enterprise developers do every day. MAI Code 2 targets deep debugging, multi-file changes, refactoring and reasoning across a whole repository. Microsoft says it aims to produce changes that code maintainers would actually merge.

MAI Code 2 Flash handles high-volume tasks. Those include explaining code, making targeted edits, fixing routine bugs and iterating quickly. GitHub describes both models as designed for agentic coding, where the model plans and carries out changes on its own.

Microsoft is also clear about the limits. The models are "explicitly not yet tuned for harder problems," such as fundamental science research and very long agentic tasks. That is the territory of frontier models like GPT-6.2 Astra and Claude Fable 5.2.

The company says MAI Code 2 is also tuned for dashboards, visualizations and other visual coding. In blind tests of front-end apps, 258 reviewers preferred its results over GPT-6 Sol's 64% of the time.

Those claims rest on a set of coding benchmarks that need a closer look.

MAI Code 2 benchmarks and how Microsoft measured them

Microsoft averaged MAI Code 2's results over four coding benchmarks. These are DeepSWE 1.1, Terminal-Bench 2.1, SWE-Atlas and FrontierCode 1.1. On that average, it reports 98% of GPT-6.1 Sol's score and 96% of Claude Opus 5.5's.

For MAI Code 2 Flash, Microsoft used three of those benchmarks and left out FrontierCode. It says Flash performs comparably to OpenAI's GPT-6 Luna at 65% of the cost and to Claude Haiku 5.5 at 20% of the cost.

The fine print matters. All models ran at maximum reasoning effort. Microsoft ran three of the benchmarks on its internal evaluation platform through the GitHub Copilot CLI, without subagents. The FrontierCode scores and costs come from Artificial Analysis.

The cost figures carry a bigger caveat. Microsoft priced both of its own models at their 50% promotional rate when it calculated the cost comparisons. At the full price that starts in May 2027, MAI Code 2 would cost about 80% of GPT-6.1 Sol rather than 40%.

Microsoft has not published a table with the raw scores, only the percentages. Independent tests have not been released yet. The pricing itself is public.

MAI Code 2 pricing in GitHub Copilot

Both models use GitHub Copilot's usage-based billing, priced per million tokens. Promotional prices run until May 17, 2027.

MAI Code 2 pricing per million tokens: $0.50 input and $2.50 output during the promotion vs $2 and $10 for GPT-6.1 Sol; Flash at $0.05 and $0.25
List prices per million tokens in GitHub Copilot and the OpenAI API, November 17, 2026; MAI prices are promotional until May 17, 2027

For comparison, GPT-6.1 Sol and the new GPT-6.2 Sol cost $2 per million input tokens and $10 per million output tokens. At full price, MAI Code 2 costs half as much per token. During the promotion it costs a quarter.

Availability depends on the Copilot plan. MAI Code 2 is open to Copilot Pro, Pro+, Max, Business and Enterprise. MAI Code 2 Flash also reaches the Free and Student plans. Developers can pick either model in VS Code, Visual Studio, JetBrains, Xcode, the Copilot CLI and on github.com.

Microsoft says the models will soon come to OpenRouter, Vercel, Hermes, OpenCode and Kilo. Code 2 Flash is already in limited preview in Microsoft Foundry. The older MAI-Code-1.1-Flash will be retired in Copilot on December 18.

The launch is one part of a wider push at Ignite.

Why Microsoft is building its own coding models

MAI Code 2 was one of several models Microsoft AI announced at Ignite. The lineup also includes MAI Cyber 2 Flash for security and the MAI Image 2.7 image editor. It adds the MAI Realtime 1 voice model and Decision 1.1, which routes requests between models. Microsoft also started offering licensed weights, so companies can run customized MAI models in their own Azure environment.

The common thread is cost. "As token spend skyrockets," Microsoft AI wrote, customers want models cheap enough to run agents around the clock. GitHub Copilot is the natural place to test that, since it is the largest coding assistant Microsoft owns. Until now, it has relied mostly on models from OpenAI, Anthropic and Google.

Microsoft is also using the same family inside its own products. A version of Code 2 Flash, called MAI Work Flash, now serves features in Microsoft 365 and Fabric IQ. Microsoft says it performs on par with GPT-6 Luna in Excel and PowerPoint tasks while using fewer tokens.

Is MAI Code 2 worth using

For everyday coding in Copilot, it is worth a try. If Microsoft's numbers hold, MAI Code 2 comes close to GPT-6.1 Sol for a quarter of its token price during the promotion. MAI Code 2 Flash adds a cheap option for routine edits, including on the free plan.

The caveats are real. The benchmarks come from Microsoft, the cost advantage halves when the promotion ends, and the models are not meant for the hardest tasks. Teams with long, complex agent jobs should keep a frontier model in the loop. Our guide to the best AI model for coding compares them by task. For the bulk of daily coding work, MAI Code 2 gives Copilot users a much cheaper default to test.

FAQ

Here are short answers to common questions about the launch.

What is MAI Code 2?

MAI Code 2 is Microsoft AI's coding model for everyday enterprise software work, released on November 17, 2026. It is available in GitHub Copilot alongside a smaller MAI Code 2 Flash.

How much does MAI Code 2 cost?

Until May 17, 2027, MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens in GitHub Copilot. After that, the price doubles to $1 and $5.

Is MAI Code 2 free in GitHub Copilot?

MAI Code 2 needs a paid Copilot plan. MAI Code 2 Flash is also available on Copilot Free and Student plans.

Sources

Portrait of Felix Laurent, author at thehype.news

Felix Laurent

News Writer at thehype. Felix is a journalist with experience in marketing. He covers artificial intelligence and emerging technologies, focusing on the ideas, innovations, and trends shaping the future of tech.

More articles by Felix Laurent →
On air · radio.thehype.news ↗ AI news radio — 24/7