Microsoft's MAI Code 2 Models Bring In-House Coding AI to Copilot
Microsoft launches MAI Code 2 and MAI Code 2 Flash at Ignite 2026. Benchmarks, per-task costs and promo pricing ending May 17, 2027 explained.
Microsoft has launched MAI Code 2 and MAI Code 2 Flash, two in-house coding models introduced at Ignite 2026 on November 17. The full MAI Code 2 model targets multi-file changes, deep debugging and repository-wide reasoning, while Flash handles high-volume routine work. Microsoft says both models compete with offerings from OpenAI and Anthropic, and both are available in GitHub Copilot.
The launch extends Microsoft's effort to build proprietary model capacity and reduce its dependence on outside providers. Microsoft states that the models were not yet tuned for fundamental science research or long-horizon agentic work, so the company positions them for everyday enterprise coding instead.
How MAI Code 2 Compares on Benchmarks
On Terminal-Bench 2.1, MAI Code 2 scored 84.6%, ahead of Claude Opus 5.5 at 78.7% but behind GPT-6.1 Sol at 85%.
The gap is widest on the harder tasks. On FrontierCode v1.1 the model scored 52%, against 56% for Claude Opus 5.5, and on SWE-Atlas it reached 43.1%, trailing Claude Opus 5.5 at 57.5% by 14 points. Developers working on long, hard repository tasks should treat those results as the current limit of the model.
MAI Code 2 Flash scored 61% on average across DeepSWE 1.1, Terminal-Bench 2.1 and SWE Atlas at maximum reasoning. According to Microsoft's comparison, that puts it above GPT-6 Luna at 55.1% and Claude Haiku 5.5 at 53.9%. Flash also beat the full MAI Code 2 on SWE Atlas, 44.8% to 43.1%, so the smaller model is not just a cheaper stand-in for every task.
What MAI Code 2 Costs Developers
Microsoft put the average cost per task at $0.309 for MAI Code 2, compared with $0.761 for GPT-6.1 Sol and $6.306 for Claude Opus 5.5. For Flash, the figure is $0.052, against $0.080 for GPT-6 Luna and $0.319 for Claude Haiku 5.5. Those cost comparisons use 50% promotional pricing.
Per GitHub's pricing page, MAI Code 2 costs $0.50 for every million input tokens, $0.025 for every million cached input tokens and $2.50 for every million output tokens until May 17, 2027. After that date the rates rise to $1.00, $0.05 and $5.00 respectively. At list price, Microsoft's own benchmark would put the cost per task at roughly 81% of GPT-6.1 Sol's cost, not the 40% implied by the promotional rate. Teams budgeting for a Copilot-based coding workflow should therefore model list prices from the start.
Each model offers a 1,050,000-token context window, which splits into up to 922,000 tokens of input and 128,000 tokens of output. Users can also choose among five reasoning effort settings, running from low to max. MAI Code 2 is available in GitHub Copilot, Visual Studio, JetBrains, Xcode, Eclipse and Copilot CLI. Flash is also available on Copilot Free and Student plans and is in private preview in Microsoft Foundry.
Where the Series Is Heading
Coming after two earlier generations, MAI Code 2 is Microsoft's newest in-house coding model. June brought MAI-Code-1-Flash to GitHub Copilot, and August added MAI Code 1.1 Flash on top of it. Microsoft also introduced six other MAI models at Ignite, and it says MAI Licensed Weights, which allow enterprise fine-tuning inside customers' own Azure environments, are coming soon.
For teams that already run GitHub Copilot, the choice is now between a cheaper in-house default for routine edits and the established models for long, hard tasks. The promotional rate ends on May 17, 2027, when MAI Code 2 list prices double, so the cost advantage over GPT-6.1 Sol narrows on Microsoft's own numbers.
Sources
Microsoft Ignite 2026 Book of News
Ignite 2026: Building the frontier ecosystem - The Official Microsoft Blog
Photo by Barney Goodman on Unsplash
Earlier coverage
- Oct 15, 2026: Microsoft Lays Out Maia 300 Chip for Multi-Gigawatt AI Inference From 2027
- Sep 25, 2026: Microsoft's Copilot Super App Unites Chat, Code and Agents in One Hub
- Sep 14, 2026: Microsoft MAI Code of Conduct Opens to Public Review as Nadella Pushes Deliberate Pacing
Related Articles
- Microsoft Launches MAI-Image-2-Efficient for lower AI Costs
- Microsoft MAI Models Cut Copilot Costs in Excel, Outlook
- Microsoft and NVIDIA Launch MAI Model Family and Unified Agentic Stack
✔Human Verified
Researched and cross-referenced against primary sources by the bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the bytevyte editorial team.