Microsoft introduced MAI Code 2 and MAI Code 2 Flash on November 17, 2026, targeting everyday enterprise coding. The pair separates repository work and debugging from high-volume explanations, small edits and routine fixes.
What does MAI Code 2 do?
Microsoft positions MAI Code 2 for deep debugging, changes across several files, repository reasoning and interface development. In its announcement, the company says the model is designed to produce changes that maintainers would accept, rather than optimizing only for an isolated coding answer.
MAI Code 2 Flash targets shorter, more frequent tasks. Microsoft lists code explanations, targeted edits and rapid iteration among its uses. The company explicitly says these models are not yet tuned for fundamental science research or very long-horizon agentic work.
What do the coding comparisons show?
Microsoft reports that MAI Code 2 reaches 98% of GPT-6.1 Sol’s performance at 40% of its cost across four coding benchmarks. Against Claude Opus 5.5, the reported figures are 96% of performance at 5% of cost. These are averages across specific evaluations, rather than a promise about every repository.
The comparison includes DeepSWE 1.1, TerminalBench 2.1, SWE-Atlas and FrontierCode 1.1. Microsoft says the models ran at maximum reasoning effort and its own benchmark runs used GitHub Copilot CLI without subagents. Some reported MAI costs use a 50% promotional price. Teams should retain those conditions when comparing the headline ratios with their own bills.
Where can developers use the models?
Both models are available in GitHub Copilot. Code 2 Flash also has a limited-access preview in Microsoft Foundry; the announcement describes other distribution partners as forthcoming.
Franklin analysis: start with a representative code change and review the resulting diff, tests and total task cost together. A cheaper generation is useful only if the complete change meets the team’s review standard. The next useful evidence is how these models handle a team’s actual repositories and repeated maintenance work.
Teams organizing coding-agent changes can also consult AOSplane’s coding-agent workflow, which covers planning and checking agent pull requests.
