Microsoft Launches MAI Code 2 and Other In-House Models at Ignite
Microsoft's own coding models went live in GitHub Copilot on November 17, with Microsoft claiming near-frontier quality at a fraction of rival prices. Security, image, voice and routing models came with them, at varying stages of availability.
By The AI Enterprise Magazine Team4 min readNews · Weekly briefing

Key takeaways
- In Copilot now. MAI Code 2 is available on paid Copilot plans; the Flash model also reaches Free and Student users.
- Cost claims. Microsoft says Code 2 reaches 98% of GPT-6.1 Sol quality at about 40% of the price, averaged across four public benchmarks.
- Stated limits. Code 2 is not yet tuned for science research or very long-horizon agentic tasks.
- Mixed availability. Cyber 2 Flash is limited to Microsoft security products, and Realtime 1 is at the interest-registration stage.
Microsoft's in-house model effort took a bigger share of the stage at Ignite 2026. On November 17 the company introduced a set of new models from its Microsoft AI group, led by MAI Code 2 and MAI Code 2 Flash for software development, which went live in GitHub Copilot the same day. The release also included a security model, two image models, a speech-to-speech model and an updated routing model. For enterprise technology leaders, the significance is less about any single benchmark and more about Microsoft putting its own models alongside OpenAI and Anthropic inside products that many organizations already pay for.
What are MAI Code 2 and MAI Code 2 Flash?
According to the GitHub changelog, MAI Code 2 is a general-purpose coding model aimed at deep debugging, multi-file implementation, refactoring and repository-wide reasoning. MAI Code 2 Flash is described as "a lightweight, cost-efficient coding model optimized for everyday development workflows," suited to explanations, targeted edits and routine fixes.
MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise plans, while the Flash model also reaches Free and Student users. Both use usage-based billing, and administrators control access through Copilot model policies. GitHub lists availability across VS Code, Visual Studio, Copilot CLI, the cloud agent, github.com, mobile, JetBrains, Xcode and Eclipse.
How does Microsoft say they compare?
The benchmark claims are Microsoft's own, as reported by TestingCatalog, and have not been independently reproduced. Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol's quality at about 40% of the price and 96% of Claude Opus 5.5's quality at about 5% of the price, averaged across DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1. For the Flash model, Microsoft says it matches GPT-6 Luna at 65% of the price and Claude Haiku 5.5 at 20% of the price on routine edits.
Microsoft was also explicit about limits. TestingCatalog reports that MAI Code 2 is tuned for everyday enterprise work and frontend output, and is not yet tuned for science research or very long-horizon agentic tasks. That matters for teams planning to use it for multi-hour autonomous coding sessions, where the frontier models from OpenAI and Anthropic remain the reference point.
What else did Microsoft release?
Several other models were announced, with availability that varies widely. MAI Cyber 2 Flash is a security model with a one-million-token context window. TestingCatalog reports that Microsoft puts its score at 92.3% on CyberGym-E2E against 91.5% for GPT-6 Astra, at under 2% of the cost per task, with list prices of $0.10 per million input tokens and $0.50 per million output tokens. It is exclusive to Microsoft's security products. TestingCatalog says it will replace the current model in Microsoft's Codename MDASH security tool from mid-December, while the Ignite Book of News describes it as preview-only in MDASH, so the timing is not settled.
MAI Image 2.7 and MAI Image 2.7 Flash are image generation and editing models. TestingCatalog reports that Artificial Analysis ranked MAI Image 2.7 first on its image editing leaderboard, at roughly one-sixth the price of the four models behind it. Sources differ on whether the model is in preview or generally available. MAI Realtime 1 is Microsoft's first live speech-to-speech model, starting at $0.01 per audio minute, with interest registration open through MAI Labs rather than general availability.
The routing model, Microsoft-Decision-1.1, decides when a request needs a larger model. TestingCatalog reports a score of 68 on the Artificial Analysis Decision Index, up from 62 for the first version, at an unchanged $0.042 per million input tokens.
Why does this matter for enterprise model strategy?
Two points stand out. First, Microsoft now ships its own coding models directly into the developer tool with the largest enterprise footprint, priced well below the frontier models it competes with. For organizations that pay for Copilot under usage-based billing, a cheaper default model for routine edits could change cost per developer meaningfully, provided quality holds on their own codebases. Microsoft's quality comparisons are averages across public benchmarks, and teams should measure acceptance rates and review effort on real work before switching defaults.
Second, the security and routing models point to a pattern in which Microsoft uses small, specialized models inside its own products and keeps larger third-party models for harder tasks. TestingCatalog also reports a program called MAI Licensed Weights that would let enterprises customize a frontier MAI model with their own data inside their own Azure environment, though Microsoft has not published availability or pricing for it.
What should buyers check?
Administrators managing GitHub Copilot Business or Enterprise should review model policies, since new models can appear in developer pickers under default enablement. Procurement and platform teams should ask Microsoft for per-token pricing for MAI Code 2 under their agreements, and ask whether usage counts toward existing Azure commitments. Security teams evaluating MAI Cyber 2 Flash should confirm when it reaches their tenant and how it compares on their own alert data. Because several availability statements differ between Microsoft's materials and outside reporting, the safest assumption is that anything not marked generally available in the admin console is still in preview.
Frequently asked questions
What is MAI Code 2?
Microsoft AI's general-purpose coding model for debugging, multi-file implementation, refactoring and repository-wide reasoning, available in GitHub Copilot from November 17, 2026.
Which GitHub Copilot plans include MAI Code 2?
MAI Code 2 is on Pro, Pro+, Max, Business and Enterprise. MAI Code 2 Flash is also on Free and Student plans. Both use usage-based billing.
How does MAI Code 2 compare with GPT-6.1 Sol and Claude Opus 5.5?
Microsoft says it reaches 98% of GPT-6.1 Sol's quality at about 40% of the price and 96% of Claude Opus 5.5's at about 5% of the price, averaged across public benchmarks. These are vendor claims.
Sources
Filed under: Agentic AI, Data & AI Infrastructure, AI Strategy & ROI
Become a Sponsor







