Microsoft Launches Seven MAI Models at Ignite, Led by Code 2 for GitHub Copilot

Photo: Hippopx (public domain). Illustrative image.
What happened. Microsoft opened Ignite 2026 in San Francisco on 17 November with the biggest model announcement in its history: seven in-house MAI models, plus a licensing programme that lets trusted customers access the model weights themselves. Chief executive Satya Nadella framed the keynote around a “frontier ecosystem” — Copilot as the new operating system for work, Microsoft IQ as the platform for an organisation’s own intelligence, and the MAI family as the models tuned end to end for real tasks. The models are the part developers will actually touch first.
The seven models. MAI Code 2 and Code 2 Flash are the flagship agentic coding models, generally available now in GitHub Copilot — in the model picker and as part of Copilot Fusion — and in Foundry. MAI Cyber 2 Flash is the security model, co-designed with Microsoft’s Perception security system. MAI Image 2.7 and Image 2.7 Flash cover image generation and editing. MAI Realtime 1 handles live voice, MAI Transcribe 2 (Dragon Copilot Transcribe) handles transcription, and Decision 1.1 — now generally available with multimodal support — is the small model for routing, labelling and classification, which Microsoft says more than 4,000 customers have already used. MAI Licensed Weights, announced alongside, lets trusted customers get the weights through Microsoft’s Frontier Co. and host them inside private Azure tenants.
The claims are aggressive, and they are Microsoft’s. The company says Code 2 delivers 98% of GPT-6.1 Sol’s coding quality at about 40% of the cost, and 96% of Claude Opus 5.5 at 5% of the cost, averaged over four Microsoft-selected benchmarks. Cyber 2 Flash, it claims, beats GPT-6 Astra on CyberGym-E2E at under 2% of the cost per task. Image 2.7 claims first place on Artificial Analysis image editing at a fifth of the price of GPT Image 2.6. Realtime 1 starts at $0.01 per audio minute. Two caveats belong in the same paragraph: the percentages are Microsoft’s own, averaged over benchmarks it chose, and secondary coverage notes the cost figures use a promotional price — no absolute per-token price was published on the primary announcement page. Microsoft also says plainly that the models are not yet tuned for fundamental science research or very long-horizon agentic work. For how these coding assistants compare in practice, see Claude Code vs Codex vs Gemini CLI and Antigravity vs Copilot vs Claude Code.
The rest of the keynote matters too. Copilot gets Workspaces — a multiplayer form factor where people and their agents work toward a shared goal with shared context — and Autopilot, an autonomous agent that keeps work moving around the clock. Microsoft IQ externalises the system underneath Copilot as a platform for organisational intelligence, with hundreds of new connectors. Agent 365, the control plane for enterprise agents, already counts over 50 million agents. Perception, the agentic security system with red, blue and green agent teams, is now generally available. Cobalt 200 VMs are generally available, and Maia 300 — Microsoft’s custom inference chip, announced in October — is reiterated at 40% more tokens per watt than leading GPU-based systems, with production deployments from 2027.
Why it matters. Microsoft is no longer just the distributor of other labs’ models: it now ships its own frontier-adjacent family, prices it as the cheap option, and puts it inside the tool — GitHub Copilot — where most enterprise developers already work. The same keynote confirmed GPT-6.2 is coming to Foundry, so Microsoft is hedging both ways: its own models for cost, everyone else’s for capability. The test for Code 2 is not the benchmark table but whether developers keep it selected in the Copilot picker once the launch week ends.
Sources
- Microsoft — “Ignite 2026: Building the frontier ecosystem”, The Official Microsoft Blog (17 November 2026): https://blogs.microsoft.com/blog/2026/11/17/ignite-2026-building-the-frontier-ecosystem/
- Microsoft AI — “Introducing MAI Code 2 and MAI Code 2 Flash” (17 November 2026): https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/
- Microsoft AI — “Microsoft Ignite 2026” (17 November 2026): https://microsoft.ai/news/microsoft-ignite-2026/
- FreshSynapse — “GPT-6.2 Astra and Sol Are Here: OpenAI’s Frontier Model Gets 20% Cheaper”: https://freshsynapse.com/articles/gpt-6-2-astra-sol-launch-2026-11-18/
- FreshSynapse — “Claude Code vs Codex vs Gemini CLI: Which AI Coding Agent Wins?”: https://freshsynapse.com/articles/claude-code-vs-codex-vs-gemini-cli-2026-11-14/