Microsoft launches seven MAI models, undercutting rivals on coding price
Microsoft AI unveiled seven new models at Ignite on November 17, led by MAI Code 2.

· Last verified: 18 Nov 2026 (Microsoft AI, Ignite Book of News, Testing Catalog)

Summary
- Microsoft AI unveiled seven new models at Ignite on November 17, led by MAI Code 2.
- Microsoft says MAI Code 2 delivers 98% of GPT-6.1 Sol coding performance at roughly 40% of the price.
- Enterprises paying frontier rates for coding models now have an in-house alternative inside GitHub Copilot and Foundry.
The latest
Seven new models arrived at once on the opening day of Microsoft Ignite in San Francisco, spanning coding, cybersecurity, image editing, live speech, clinical transcription and request routing. Microsoft AI framed the launch around price-performance, claiming near-frontier quality at a fraction of competitors' cost. The conference runs November 17–20.
Details
- The flagship: MAI Code 2 is pitched at enterprise software teams. Microsoft says it reaches 98% of GPT-6.1 Sol's coding performance at about 40% of the price, and 96% of Claude Opus 5.5 at 5% of the price, averaged across the DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1 benchmarks.
- The caveat: Microsoft states MAI Code 2 is not yet tuned for scientific research or long-horizon agentic tasks. In a blind front-end evaluation, developers preferred it to GPT-6 Sol in 68% of cases, the company says. It is generally available in GitHub, per the Ignite Book of News.
- Where it runs: Testing Catalog, citing the GitHub changelog, reports MAI Code 2 is available to Copilot Pro, Pro+, Max, Business and Enterprise plans across VS Code, Visual Studio, the CLI, JetBrains, Xcode and Eclipse.
- The cheaper tier: MAI Code 2 Flash targets routine edits and fixes, with performance Microsoft describes as comparable to GPT-6 Luna at 65% of its price and Claude Haiku 5.5 at 20%. It reaches free and student Copilot plans, plus a limited-access preview in Microsoft Foundry.
- Security model: MAI Cyber 2 Flash matches GPT-6 Astra's quality at under 2% of the task cost on CyberGymE2E, per Microsoft, with output tokens 86% cheaper than MAI Cyber 1.1 Flash. MAI Cyber now handles about 40% of scans in MDASH, and 90% of scans started on MAI stayed on it.
- The numbers: Testing Catalog puts MAI Cyber 2 Flash at a 1 million-token context window and 92.3% on CyberGym-E2E against 91.5% for GPT-6 Astra, priced at 0.10 dollars per million input tokens and 0.50 dollars per million output tokens. It replaces the current model in Codename MDASH for eligible customers from mid-December.
- Images: Microsoft says MAI Image 2.7 ranks first in image editing on Artificial Analysis, ahead of GPT Image 2.6 Sunburst (max) at one-fifth the price, and fourth in text-to-image. MAI Image 2.7 Flash runs about 10 dollars per 1,000 images and powers the new Canvas space in MAI Labs.
- Voice and health: MAI Realtime 1 is Microsoft's first live speech-to-speech model, starting at 0.01 dollars per audio minute and available to try in MAI Labs. Dragon Copilot Transcribe, built on MAI-Transcribe-2, covers 10 million encounters monthly and is used by a third of US physicians, the company says.
- Routing and silicon: Decision 1.1, a request-routing model Microsoft says 4,000 customers use, is billed as 50 times faster and 79 times cheaper than GPT-Sol 6.1. Testing Catalog reports 68 points on AA Decision Index v2.1 against 62 for the prior version. Maia 200, the second-generation accelerator, is in production with 1.7 times more tokens per watt.
- Custom builds: Microsoft says AT&T built a custom MAI model for 5G network operations analytics that outperformed GPT-6.1 Sol at 50 times lower cost. The MAI Licensed Weights program lets customers fine-tune a frontier MAI model on their own data inside their Azure environment via Frontier Company, pitched with the line "no one wants to pour their IQ into someone else's model."
Background
Microsoft Foundry hosts more than 11,000 models, and its model router covers the top 50 of them, according to the Ignite Book of News. That distribution layer is where most of the new MAI releases land for enterprise customers.
Between the lines
Every headline claim is a cost ratio rather than an outright capability win: 40% of the price, 5% of the price, under 2% of task cost, 50 times cheaper. Microsoft is not arguing MAI beats GPT-6.1 Sol or Claude Opus 5.5 on quality, but that the remaining gap is small enough to be worth the savings. The admitted weakness in research and long-horizon agentic work marks where that argument stops.
What's next
Mid-December is when MAI Cyber 2 Flash is set to replace the existing model in Codename MDASH for eligible customers. Ignite runs through November 20, and the Foundry previews for Image 2.7 and Code 2 Flash have no announced general-availability date.