Microsoft Launches MAI Code 2 to Challenge OpenAI and Anthropic API Costs

Microsoft Launches MAI Code 2 to Challenge OpenAI and Anthropic API Costs

At its Ignite conference, Microsoft introduced MAI Code 2 and MAI Code 2 Flash, two proprietary AI models for everyday software development. Internal testing shows the flagship model delivers 96 % of Claude Opus 5.5 performance at 5 % of the API cost, and matches 98 % of GPT 6.1 Sol at 40 % of the price. Both models are available in GitHub Copilot, though Microsoft notes they are not tuned for long horizon agentic work or fundamental science research.

Microsoft evaluated MAI Code 2 on four enterprise coding benchmarks DeepSWE 1.1, Terminal Bench 2.1, SWE Atlas, and FrontierCode v1.1. The results highlight a substantial cost reduction for companies that pay per API call rather than using flat consumer subscriptions. The model handles production level code, deep debugging, multi file edits, and repository wide reasoning, and performs well in visual interface tasks. In blind frontend evaluations, developers preferred apps built by MAI Code 2 over GPT 6 Sol 64 % of the time.

The lighter MAI Code 2 Flash targets routine programming. Across three internal benchmarks covering code explanations, targeted edits, and quick fixes, it matches GPT 6 Luna at 65 % of the price and delivers comparable output to Claude Haiku 5.5 at 20 % of the cost. Building these models in house reduces Microsoft’s reliance on OpenAI and Anthropic APIs.

A version for productivity suites, MAI Work Flash, adapts Code 2 Flash for Microsoft 365 and FabricIQ. It matches GPT 6 Luna on accuracy and token efficiency in standard Excel and PowerPoint tasks, and shows greater savings in Word. Across five internal benchmarks, MAI Work Flash reaches GPT 6.1 Sol quality while cutting input tokens by roughly 42 % per call. Within Teams, it outperforms GPT 6 Luna in tracking decisions and following custom instructions, using up to 28 % fewer tokens.

Power BI’s data analytics features also receive an upgrade via FabricIQ. Offline runs show the MAI model surpassing GPT 5.6 Sol, GPT 6 Astra, and Opus 5.5 in generating DAX queries from natural language chats.

Developers can access MAI Code 2 and Code 2 Flash in GitHub Copilot today. Code 2 Flash is also available in a limited access preview on Microsoft Foundry, with releases planned for OpenRouter, Vercel, Hermes, OpenCode, and Kilo.

About the author

mgtid
mgtid
Owner of Technetbook | 10+ Years of Expertise in Technology | Seasoned Writer, Designer, and Programmer | Specialist in In-Depth Tech Reviews and Industry Insights | Passionate about Driving Innovation and Educating the Tech Community Technetbook

Join the conversation

Newsletter Subscription