Back home

MAI Code 2 / MAI Code 2 Flash

sitemap+web 0.95
Visit source

Intro

MAI Code 2 and MAI Code 2 Flash are the coding models Microsoft AI introduced on 17 November 2026 (tagged Ignite · Models). They are positioned for everyday enterprise coding at a lower cost, not for scientific research or very long-horizon agentic work. Officially, Code 2 is aimed at production-ready changes, deep debugging, multi-file implementation, repo-wide reasoning, and UI; Flash is aimed at high-volume work such as explanations, targeted edits, routine fixes, and rapid iteration.

In the same Ignite model suite, this radar already filed MAI Image 2.7 / MAI Image 2.7 Flash in the 04:32 slot. Code 2 and Code 2 Flash were not yet in the database; the official news page’s sitemap lastmod falls in this window. Microsoft also says it further optimized a single model from Code 2 Flash, MAI Work Flash, to serve multiple Microsoft 365 and Fabric IQ products. That name is treated here as an in-product variant, not a separate filed term.

The post says both models are available in GitHub Copilot, Code 2 Flash is in a limited-access preview in Microsoft Foundry, and they will follow on OpenRouter, Vercel, Hermes, Open Code, and Kilo. Customization is pointed to Microsoft AI’s Custom Models / Licensed Weights pages.

Advertisement

Explain

The published comparisons should be read with the footnotes. Microsoft says that, averaged over four benchmarks aimed at everyday GitHub Copilot enterprise tasks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1), MAI Code 2 reaches about 98% of GPT-6.1 Sol’s performance at about 40% of the cost, and about 96% of Claude Opus 5.5’s performance at about 5% of the cost. In a blind frontend evaluation covering UIs, games, visualizations, 3D scenes, and SVG, reviewers preferred Code 2 over GPT-6 Sol 64% of the time, with 258 annotators. Flash is described as comparable to GPT-6 Luna and Claude Haiku 5.5 at about 65% and 20% of the cost on three high-volume routine coding benchmarks. Footnotes state that some costs are shown at a 50% promotional price, and that results vary by task, reasoning effort, and pricing assumptions.

A “Beyond coding” section reports further-optimized internal results in Excel, PowerPoint, Word, Teams, and Fabric IQ (including DAX generation for Power BI semantic models), under the MAI Work Flash name. Those are in-product evaluations and do not mean Work Flash is sold as a separate public model. Microsoft also says the two coding models are “explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work.”

Usage

According to the official getting-started note, use MAI Code 2 and Code 2 Flash in GitHub Copilot; Code 2 Flash is also in a limited-access preview in Microsoft Foundry. Microsoft says they will also be released on OpenRouter, Vercel, Hermes, Open Code, and Kilo. For adapting a model to your own coding standards or workflows, the same article points to the Custom Models page and the Licensed Weights path.

Officially, reach for Code 2 when you want mergeable production changes, repo-wide reasoning, or stronger visual/frontend work, and for Flash when the load is explanations, small edits, and rapid iteration. Promotional pricing, preview eligibility, and later channels are those shown by Foundry, Copilot, and partners at the time of use. Do not treat MAI Work Flash as a separately shipped public model, and do not confuse this entry with MAI Image 2.7, which was already filed in the 04:32 slot.