
Microsoft used its Ignite stage on Tuesday, November 17 to launch not one model but a whole family of them — seven new first-party MAI models covering coding, cybersecurity, images, voice, transcription and automated decisions. The pitch is not raw power. It is price: Microsoft says specialized models can get you almost all of the frontier quality for a small fraction of the frontier bill, and the two coding models are already live in GitHub Copilot today.
In its announcement, Microsoft frames the strategy plainly: “Different jobs call for a different balance of capability, speed and cost. Today’s new models give you more options to match each one.” The models join a Microsoft Foundry catalog that already lists more than 11,000 models from Microsoft, OpenAI, Anthropic, Meta and others.
MAI-Code-2: the headline act
MAI-Code-2 is built, in Microsoft’s words, for “demanding agentic coding work, such as multi-step tasks, complex refactors and agent-driven development,” while MAI-Code-2-Flash is tuned for high-volume everyday coding. Both are available now in GitHub — Code-2 on Copilot Pro, Pro+, Max, Business and Enterprise plans, Flash on those plans plus Free and Student — with Flash also in limited-access preview in Foundry.
The numbers Microsoft is quoting are aggressive: it says Code-2 delivers about 98% of GPT-6.1 Sol’s coding quality at roughly 40% of the price, and 96% of Claude Opus 5.5’s performance at just 5% of the cost, averaged across benchmarks including DeepSWE 1.1 and Terminal Bench 2.1. Treat those as Microsoft-reported figures — and note the fine print other outlets flagged: the Code-2 cost math includes a 50% promotional price. Even so, the timing is pointed: OpenAI launched its own cheaper pair the very same day — see our coverage of GPT-6.2 Astra and Sol.
A security model, an image model and a voice
MAI-Cyber-2-Flash is purpose-built for cybersecurity, spanning vulnerability discovery, validation and remediation, and was trained inside Microsoft’s Codename MDASH harness, through which it will be offered to verified defenders. Microsoft says it matches GPT-6 Astra’s quality on the CyberGym-E2E benchmark at under 2% of the cost per task — a claim that lands in the same week Google documented an AI that turns patches into exploits on its own, which we covered here. MAI-Image-2.7 and its Flash sibling are in public preview in Foundry, with Microsoft claiming the top spot on Artificial Analysis image editing at a fifth of the price of the model behind it. The family rounds out with MAI Realtime 1, Microsoft’s first live speech-to-speech model, Dragon Copilot Transcribe, and the general availability of Microsoft-Decision-1.1, a small Gemma-based model for routing and classification.
Why it matters
The model war is shifting from “biggest brain wins” to “cheapest good-enough brain wins.” If Microsoft’s numbers hold up outside its own benchmarks, teams running coding and security agents around the clock could cut their model bills dramatically — and Microsoft, not OpenAI or Anthropic, would collect the savings pitch. Expect independent benchmarks to test that 96%-for-5% claim within days.
Source: Microsoft — “From model choice to your durable advantage” (Ignite 2026, Microsoft Tech Community). Performance and price comparisons are Microsoft-reported.
- What Happens If You Use a Gaming GPU for AI? The RTX 5090 Deep Dive - November 18, 2026
- What Happens If You Game on a $30,000 AI GPU? It Loses to a Laptop - November 18, 2026
- The Ultimate $10,000 Gaming PC Build in 2026: RTX 5090, 9950X3D and Zero Compromises - November 18, 2026