MAI Code 2 and MAI Code 2 Flash are coding models aimed at everyday enterprise work, with an emphasis on production-ready results at lower cost than competing models. Code 2 is built for deeper tasks such as debugging, implementing changes across multiple files, reasoning across repositories, and creating user interfaces. Flash targets frequent, lighter jobs - including explaining code, making focused edits, fixing routine issues, and supporting fast iteration. Neither is yet optimized for fundamental scientific research or long-running agentic tasks. Across four coding benchmarks, Code 2 delivers 98% of GPT-6.1 Sol’s performance at 40% of its cost and 96% of Claude Opus 5.5’s performance at 5% of its cost. In blind evaluations of visual coding, reviewers preferred Code 2’s results to GPT-6 Sol’s 67% of the time. Flash performs comparably to GPT-6 Luna and Claude Haiku 5.5 on routine coding benchmarks, at 65% and 20% of their respective costs.
The models also underpin MAI Work Flash, a version adapted for Microsoft products. Internal evaluations report performance on par with GPT-6 Luna for selected Excel and PowerPoint tasks, and similar to GPT-6.1 Sol in Word while using about 42% fewer input tokens per call. In Teams, it exceeds GPT-6 Luna on several instruction-following and tool-use tasks while using up to 28% fewer tokens; in Fabric IQ, it leads several compared models on generating DAX queries for Power BI. Code 2 and Flash are available in GitHub Copilot, while Flash is in limited-access preview in Microsoft Foundry. Benchmark results depend on task, configuration, reasoning effort, and pricing assumptions; some cost comparisons use promotional pricing.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.