ModelsNews
Microsoft MAI Code 2: in-house coding model in Copilot
At Ignite 2026, Microsoft announced seven new in-house MAI models, including the MAI Code 2 and Code 2 Flash coding models, which are already live in GitHub Copilot. Price is the main pitch: by Microsoft's own measurements, Code 2 delivers 98% of GPT-6.1 Sol's performance at about 40% of the cost.
· 3 min read

What shipped?
Microsoft AI introduced seven models on November 17 (Microsoft AI). The rationale is simple: token spend has exploded, and enterprises want a price point that lets them run agents around the clock on real engineering work. The lineup:
- MAI Code 2: general coding model for debugging, multi-file implementation and repo-wide work.
- MAI Code 2 Flash: smaller, cheaper model for routine fixes, explanations and fast iteration.
- MAI Cyber 2 Flash: defensive security model, 86% cheaper on output tokens than its predecessor.
- MAI Image 2.7: Microsoft says it ranks first on Artificial Analysis image editing.
- MAI Realtime 1: its first live speech-to-speech model, from $0.01 per audio minute.
- Dragon Copilot Transcribe (built on MAI Transcribe 2) and Decision 1.1 for model routing.
What can MAI Code 2 do?
According to the model card, both models are sparse MoE with an integrated vision encoder and a 1.05M-token context window (model card). Code 2 has 77 billion active and roughly 2.8 trillion total parameters; Flash has 12 billion active and about 398 billion total. Reinforcement learning ran directly inside the GitHub Copilot CLI and VS Code agent harnesses, so gains carry over to real workflows.
Microsoft reports an average over four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, FrontierCode v1.1) (Microsoft AI). By that measure Code 2 reaches 98% of GPT-6.1 Sol at about 40% of the cost, and 96% of Claude Opus 5.5 at 5% of the cost. In blind frontend comparisons, developers preferred Code 2 over GPT-6 Sol 64% of the time.
Two caveats from the fine print. Cost figures use the 50% promotional price of Code 2 and Flash. Microsoft also states the models are explicitly not yet tuned for hard scientific problems or very long-horizon agentic work. This is a model for everyday enterprise coding, not a frontier contender.
Where can you use it?
In GitHub Copilot, Code 2 is available on Pro, Pro+, Max, Business and Enterprise; Code 2 Flash also on Free and Student (GitHub). Both appear in the model picker in VS Code, Visual Studio, JetBrains IDEs, Xcode, Copilot CLI and the cloud agent. Code 2 Flash is in limited preview in Microsoft Foundry, with OpenRouter and Vercel availability promised later.
Why it matters
Microsoft has long leaned on OpenAI's models; now it is running its own models inside its own Copilot products, and says MAI Work Flash, optimized from Code 2 Flash, already powers M365 and Fabric features. Teams using Copilot should try Code 2 Flash on routine tasks and keep expensive models for the hard cases. Together with the newly released GPT-6.2 Astra and Sol, the price war is clearly pushing the market toward cheaper tokens.
Key takeaways
- Seven new MAI models at Ignite: coding, security, image, voice, decisions.
- MAI Code 2: 77B active parameters, 1M context, live in GitHub Copilot.
- Vendor numbers: 98% of GPT-6.1 Sol at about 40% of the cost, at promo pricing.
- Not tuned for hard science or very long-horizon agentic work.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About


