Advertise
AI

Microsoft Unveils MAI Code 2 Coding Model at Ignite 2026

Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol's coding benchmark performance at 40% of the cost, based on its own evaluation.

Microsoft unveiled a family of new in-house AI models at its annual Ignite 2026 event in San Francisco on November 17, led by the coding model MAI Code 2. The company also introduced the lightweight MAI Code 2 Flash and the security-focused MAI Cyber 2 Flash, and officially released Microsoft-Decision-1.1, a model specialized for AI agent decision tasks that now adds image input.

MAI Code 2 is a mixture-of-experts model with 2.8 trillion total parameters, activating 80 billion during inference. It supports a context of up to 1.05 million tokens and accepts image input. Microsoft says the model was pretrained on about 50 trillion tokens and then reinforced in GitHub Copilot's real development environment.

According to Microsoft's own evaluation, MAI Code 2 reached 98% of GPT-6.1 Sol's performance at 40% of its cost, averaged across four coding benchmarks. Against Anthropic's Claude Opus 5.5, the company says it reached 96% of the performance at 5% of the cost. In a blind evaluation with 258 raters, 67% preferred MAI Code 2's output over GPT-6 Sol for UI, games and data visualization.

The smaller MAI Code 2 Flash has 398 billion total parameters, activates 12 billion, and supports the same 1.05 million token context. Microsoft says it approaches the coding performance of GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their cost respectively. MAI Code 2 and Flash are available in GitHub Copilot, and Flash is also offered in limited form on Microsoft Foundry.

Security model and licensing

MAI Cyber 2 Flash scored 92.3% on the CyberGym-E2E vulnerability detection and reproduction benchmark, against 91.5% for GPT-6 Astra, at under 2% of the cost per task. Microsoft will integrate it into its security agent system for some customers from mid-December. The company said more than 4,000 customers have already used Microsoft-Decision-1.1, and announced an MAI Licensed Weights strategy that lets enterprise customers license its model weights for customization.

Quick answers

Where can I use MAI Code 2?

MAI Code 2 and MAI Code 2 Flash are available in GitHub Copilot, and Flash is also offered in limited form on Microsoft Foundry.

How large is MAI Code 2?

It is a mixture-of-experts model with 2.8 trillion total parameters, activating 80 billion during inference, with a context of up to 1.05 million tokens.

When does MAI Cyber 2 Flash reach customers?

Microsoft will integrate it into its security agent system for some customers from mid-December.

Sources