On November 17, 2026, Microsoft unveiled seven new MAI models at Ignite 2026, covering coding, image generation, real-time voice, and cybersecurity, with a single selling point: capability close to the flagship, at a fraction of the price. MAI Code 2 reaches 98% of GPT-6.1 Sol's quality across four enterprise coding benchmarks, at about 40% of the price; against Claude Opus 5.5, it delivers 96% of the performance for just 5% of the price.
Seven models, seven jobs
The seven are MAI Cyber 2 Flash, MAI Code 2, MAI Code 2 Flash, MAI Image 2.7, MAI Realtime 1, Dragon Copilot Transcribe, and Decision 1.1. Security model MAI Cyber 2 Flash beats GPT-6 Astra on quality in the CyberGymE2E benchmark, at less than 2% of the cost per task, and its output tokens cost 86% less than the previous generation. Image model MAI Image 2.7 tops the Artificial Analysis image-editing leaderboard, outperforming the full-strength GPT Image 2.6 Sunburst at one-fifth of the price; the high-volume version costs about $10 per 1,000 images. Real-time voice model MAI Realtime 1, Microsoft's first speech-to-speech model, beats GPT-Realtime 2.1 in evaluations and starts at $0.01 per audio minute.
How to use them, where they run
MAI Code 2 is available in GitHub Copilot starting today; Code 2 Flash joins Copilot and a limited preview in Microsoft Foundry, while Image 2.7 and Realtime 1 are available through Foundry and MAI Labs. Microsoft also launched MAI Licensed Weights: enterprises can bring a frontier model into their own Azure environment, customize it with their own data, control who can access it, and keep their data and adaptations in their own hands. The second-generation in-house accelerator Maia 200 is also in production, delivering 1.7x more tokens per watt.
Microsoft's play is blunt: instead of chasing OpenAI and Anthropic on peak capability, it is pushing 'good enough and cheap' to the limit, aimed squarely at the enterprise bill for agents running around the clock. For developers, Copilot gains an affordable everyday coding option; the real test comes later — whether the low-cost models hold up in real production, and whether big customers buy into Licensed Weights. When OpenAI launched GPT-6.2, it made a similar move on price, GPT-6.2 Astra and Sol Are Here: Astra Costs 20% Less, Sol Nearly Matches the Flagship — the price war between the two is now out in the open.