Microsoft used its Ignite 2026 conference in San Francisco on Tuesday to launch seven first-party MAI models at once — covering coding, cybersecurity, images, voice and transcription — and it is selling every one of them the same way: near-frontier quality at a fraction of what OpenAI and Anthropic charge. Why it matters: Microsoft is no longer just reselling other labs’ models; it is now competing with its own suppliers on price.
November 17, 2026 — San Francisco
What happened
Microsoft AI, the division led by Mustafa Suleyman, announced the new model family at Ignite on November 17. Reuters reports the strategy plainly: build models tuned for specific domains so they stay cheap to run while matching — and on some benchmarks topping — the big general-purpose rivals. Alongside the seven MAI models, Microsoft made its Microsoft-Decision-1.1 routing model generally available and launched MAI Licensed Weights, which lets enterprises customize a frontier MAI model with their own data and run it inside their own Azure environment.
The lineup
| Model | Job | Status | Microsoft’s headline claim |
|---|---|---|---|
| MAI Code 2 | Agentic coding | Generally available in GitHub Copilot | 98% of GPT-6.1 Sol’s quality at about 40% of the price |
| MAI Code 2 Flash | Everyday coding | Live in Copilot; limited preview in Foundry | Matches GPT-6 Luna at 65% of the price |
| MAI Cyber 2 Flash | Cybersecurity | Coming to Microsoft security products | 92.3% on CyberGym-E2E vs 91.5% for GPT-6 Astra, at under 2% of the cost per task |
| MAI Image 2.7 | Image generation and editing | Public preview | #1 on Artificial Analysis image editing, at about $35 per 1,000 images |
| MAI Image 2.7 Flash | Fast, cheap image work | Public preview; powers the new Canvas workspace in MAI Labs | About $10 per 1,000 images |
| MAI Realtime 1 | Live speech-to-speech | Interest registration via MAI Labs | Microsoft’s first speech-to-speech model, from $0.01 per audio minute |
| MAI Transcribe 2 | Transcription | Behind Dragon Copilot Transcribe | Already handling 10 million patient encounters a month |
All benchmark figures above are Microsoft’s own numbers. The Code 2 comparisons are averaged across DeepSWE 1.1, Terminal-Bench 2.1, SWE-Atlas and FrontierCode v1.1, and Microsoft is candid that Code 2 is “not yet tuned” for fundamental science research or very long agentic tasks. GitHub’s changelog puts Code 2 in Copilot for Pro, Pro+, Max, Business and Enterprise plans, with Code 2 Flash also reaching Free and Student plans.
The details worth knowing
Cyber is the guarded one. MAI Cyber 2 Flash carries a one-million-token context window and token pricing of $0.10 input / $0.50 output per million tokens, but it is not going on general sale: Reuters says Microsoft plans to offer it to a limited group of customers, and the model is slated to replace the current model inside Codename MDASH — Microsoft’s agentic code-security system — for eligible customers from mid-December.
Images are the bragging right. Artificial Analysis independently ranks MAI Image 2.7 first on its image-editing leaderboard, ahead of GPT Image 2.6 Sunburst (max), at roughly a fifth of the price. Both image models support outputs up to 2,048 × 2,048 pixels and transparent backgrounds, and are in public preview through Microsoft Foundry and partners including fal, Genspark, OpenRouter and Vercel.
Routing is the quiet money-saver. Microsoft-Decision-1.1, now generally available at $0.042 per million input tokens, decides which requests actually need a big model. Artificial Analysis scores it 68 on its Decision Index, up from 62 for the previous version, and SiliconANGLE notes the model is built on a new base — Google’s Gemma 4 26B A4 — with multimodal support. Microsoft also says its Maia 200 accelerator is now in production running its largest models.
Why it matters
For subscribers, the practical read is this: the price of “good enough” AI just dropped again. If Microsoft’s claims hold up in real use, a coding model at 40% of GPT-6.1 Sol’s price — already sitting inside GitHub Copilot, where many developers already pay — pressures every rival’s pricing, including the OpenAI models Microsoft itself sells through Azure. The cyber model shows the other half of the strategy: the most sensitive capabilities stay gated to vetted customers inside Microsoft’s own security products. Everything AI, tested — we will be watching whether independent results match the launch-day charts.
FAQ
Can I use MAI Code 2 today?
Yes, if you have a paid GitHub Copilot plan (Pro and up); Code 2 Flash is also available on Copilot’s Free and Student plans. Code 2 Flash is additionally in a limited preview in Microsoft Foundry.
How much does MAI Image 2.7 cost?
In public preview on Microsoft Foundry, about $35 per 1,000 images for the full model and about $10 per 1,000 for Flash, per Microsoft’s announcement.
Can anyone buy the MAI Cyber 2 Flash model?
No. Microsoft is limiting it to eligible customers inside its own security products, starting with Codename MDASH in mid-December.
Sources: Reuters, SiliconANGLE, Microsoft AI, GitHub Changelog, Artificial Analysis

