Microsoft strengthens its MAI range for businesses at a competitive price

Microsoft unveiled seven MAI models at Ignite 2026, focusing on code, security, image, voice, and routing. The goal is clear: to offer near high-end performance at a lower cost for enterprise use.

At Ignite 2026, Microsoft introduced a new family of seven MAI models aimed at businesses deploying agents continuously and monitoring every token bill like a very diligent treasurer. The range covers code, cybersecurity, image, voice, transcription, and request routing. The challenge is simple: to deliver results close to the very high end without paying a premium for each model call.

Models designed for coding and everyday tasks

On the development side, MAI Code 2 and MAI Code 2 Flash are the highlights. Microsoft claims that Code 2 reaches 98% of the quality of GPT-6.1 Sol for about 40% of the price, and 96% of Claude Opus 5.5 for 5% of the price, on benchmarks like DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1. The model mainly targets daily software work and front-end interfaces, but Microsoft notes it is not yet optimized for scientific research or very long agentic tasks.

Code 2 Flash is more suited for quick fixes and minor tweaks. Again, Microsoft promotes an aggressive pricing position: the model is said to be at the level of GPT-6 Luna at 65% of the price, and Claude Haiku 5.5 at 20%. Both are available in GitHub Copilot, while Flash is also in limited preview on Microsoft Foundry.

Sécurité, image et voix : Microsoft élargit le terrain

The cybersecurity segment relies on MAI Cyber 2 Flash. Microsoft says the model scores 92.3% on CyberGym-E2E, compared to 91.5% for GPT-6 Astra, at less than 2% of the cost per task. It also features a one million token context window, a capacity that allows ingesting very large volumes of text. Its announced price is 0.10 dollars input and 0.50 dollars output per million tokens. The model is expected exclusively in Microsoft security products, with a planned replacement starting mid-December for eligible Codename MDASH customers.

On the image side, MAI-Image-2.7 and MAI-Image-2.7-Flash position themselves on quality and price. Microsoft indicates that MAI-Image-2.7 takes first place on Artificial Analysis in image editing, ahead of GPT Image 2.6 Sunburst, for about one fifth of the price. The Flash version would cost about 10 dollars for 1,000 images, compared to about 35 dollars for the full model. Both are in public preview on Foundry and via partners like fal, Genspark, OpenRouter, and Vercel. Flash also powers a new Canvas space in MAI Labs.

Voice and transcription complete the picture. MAI Realtime 1 is presented as Microsoft's first live speech-to-speech model, with a starting price set at 0.01 dollars per audio minute and open registration via MAI Labs. Meanwhile, Dragon Copilot Transcribe, built on MAI Transcribe 2, already processes 10 million patient encounters per month.

Le routage et le déploiement privé, la pièce discrète mais stratégique

Microsoft also launched Microsoft-Decision-1.1, a routing model that only sends requests to large models when necessary. This type of tool helps reduce costs and avoid mobilizing a large model for a routine request. Version 1.1 is generally available and scores 68 on the Artificial Analysis Decision Index, compared to 62 for the previous version. Microsoft also announces MAI Licensed Weights, which allows a company to customize a frontier MAI model with its own data and run it in its Azure environment.

Finally, the company says its Maia 200 accelerator is now in production for its largest models. The commercial message is clear: Microsoft wants to become a provider of cheaper models, yet powerful enough to fuel Copilot, Foundry, and Azure. The competition with OpenAI and Anthropic is therefore no longer only about raw quality; it also plays on the performance-price ratio, a field where cost sometimes matters as much as the benchmark.

Key points

  • Seven MAI models were announced at Ignite 2026.
  • MAI Code 2 is integrated into GitHub Copilot.
  • MAI Cyber 2 Flash targets Microsoft security products.
  • MAI Realtime 1 starts at 0.01 dollars per audio minute.
  • Microsoft-Decision-1.1 is generally available.
  • MAI Image 2.7 is in public preview on Foundry.

By the numbers

  • 98% — claimed quality of MAI Code 2 compared to GPT-6.1 Sol, according to Microsoft.
  • 92.3% — score of MAI Cyber 2 Flash on CyberGym-E2E, according to Microsoft.
  • 1 million tokens — context window of MAI Cyber 2 Flash, according to Microsoft.
  • 10 million — patient encounters processed per month by Dragon Copilot Transcribe.
  • 68 — score of Microsoft-Decision-1.1 on the Artificial Analysis Decision Index.

To read