Microsoft AI unveiled seven models at its Ignite conference on November 17, and the one developers can use today is MAI Code 2, which is already in GitHub Copilot. Microsoft AI, the company’s in-house model group, says the model delivers 98% of GPT-6.1 Sol’s coding quality for about 40% of the cost. Against Anthropic’s Claude Opus 5.5, it claims 96% of the performance for 5% of the price.
The pitch is price. Dividing Microsoft’s own numbers, quality per dollar works out to roughly 2.5 times GPT-6.1 Sol’s, and the Opus comparison is far steeper. Those figures are averages across four benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1. Microsoft picked them because they resemble the everyday enterprise tasks it sees in GitHub Copilot. The announcement on microsoft.ai includes no independent results.
Microsoft is also upfront about what the model skips. It is not yet tuned for fundamental science research or very long horizon agentic work. Microsoft also ran a blind comparison on frontend work: developers chose MAI Code 2 over GPT-6 Sol 64% of the time, the company says.
MAI Code 2 Flash handles the high-volume end, meaning explaining code, small edits, routine bug fixes, and fast iteration. Microsoft says it matches GPT-6 Luna at 65% of the price and Claude Haiku 5.5 at 20% of the price, measured on three benchmarks. Copilot has it now, while Microsoft Foundry offers a restricted preview.
The rest of the lineup reaches well beyond code. MAI Cyber 2 Flash, Microsoft says, scores above GPT-6 Astra on CyberGymE2E while charging under 2% as much for each task. Earlier MAI Cyber models handle almost 40% of scans in MDASH, and 90% of scans that started with MAI stayed with it. The new model’s output tokens cost 86% less than MAI Cyber 1.1 Flash.
On images, MAI Image 2.7 holds the top spot on Artificial Analysis for image editing, ahead of GPT Image 2.6 Sunburst (max) at one fifth of the price. In text-to-image it ranks fourth, which Microsoft frames as second among labs. A Flash version costs about $10 per 1,000 images, and both sit in Foundry.
MAI Realtime 1 debuts as the company’s opening speech-to-speech model for live conversation. Microsoft says it bested GPT-Realtime 2.1 on external and Dynamics 365 benchmarks, and prices begin at $0.01 for each audio minute. Access runs through MAI Labs.
Three more pieces round it out. Dragon Copilot Transcribe, built on MAI Transcribe 2, already covers 10 million patient encounters monthly, halves multilingual error rates, and sits with a third of US physicians. Decision 1.1, a model router used by 4,000 customers, runs 50x faster than GPT-6.1 Sol at a cost 79x lower, per Microsoft. Maia 200, the second-generation accelerator, is in production and delivers 1.7x the tokens for each watt.
Microsoft also introduced MAI Licensed Weights. Customers take a frontier MAI model, tune it on proprietary data, and run it inside their own Azure environment. AT&T built a custom model for 5G network analytics that Microsoft says beat GPT-6.1 Sol at 50x lower cost.
Every comparison here is a vendor result on a vendor-selected benchmark set, so the 98% is a claim to test, not a settled fact. Teams with Copilot seats can run MAI Code 2 on their own repositories this week; a result that holds on their code is worth far more than Microsoft’s average.
Microsoft AI (microsoft.ai), announcement published November 17, 2026.