Microsoft Prices MAI Code 2 at 40 Percent of GPT-6.1 Sol and Starts Licensing the Weights
Microsoft used Ignite 2026 on November 18 to launch seven MAI models and a new way to buy them. MAI Code 2 averages 98% of GPT-6.1 Sol’s quality at 40% of the cost and 96% of Claude Opus 5.5’s at 5%, measured over DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode 1.1 on Microsoft’s own MSBench harness, and developers preferred it to GPT-6 Sol in 64% of 258 blind frontend comparisons. The cost figures assume a 50% promotional price, which doubles them at list.
The rest of the batch, with the numbers Microsoft published:
- MAI Code 2 Flash, the high-volume variant, matches GPT-6 Luna at 65% of the cost and Haiku 5.5 at 20%.
- MAI Cyber 2 Flash scores 92.3% on CyberGym-E2E against 91.5% for GPT-6 Astra at under 2% of the cost per task, with a 1 million token context. Microsoft dropped 92 of the benchmark’s 905 tasks after finding they drew on the same open source projects and vulnerability reports as its training data, and reports the score on the remaining 813.
- Microsoft-Decision-1.1 scores 68 on the Artificial Analysis Decision Index v2.1, six points above the first version and ahead of Gemini Decision 1 and GPT-6 Luna, at $0.042 per million tokens.
MAI Image 2.7, MAI Realtime 1 and Dragon Copilot Transcribe fill out the list. Separately, MAI Licensed Weights lets a business fine-tune MAI models and run them inside its own Azure environment, available through Frontier Company engagements. Microsoft states plainly that Code 2 is not tuned for fundamental science research or very long-horizon agentic work.
Related: Microsoft Releases the Decision 1 Model, Microsoft Ships MAI Cyber 1.1 Flash Inside MDASH
Introducing MAI Code 2 · MAI Cyber 2 Flash · MAI Licensed Weights · Source channel