Microsoft ships its own AI models, and undercuts OpenAI on price
At its Build conference, Microsoft stopped reselling other people's AI and started shipping its own. The MAI family spans coding, reasoning, voice and image, and runs on Microsoft's Azure data centres, which lets it cut the price. MAI-Code-1-Flash is live now in GitHub Copilot and VS Code. The reasoning model, MAI-Thinking-1, is in private preview through Microsoft Foundry. The headline claim, from Microsoft AI chief Mustafa Suleyman: after tuning for McKinsey's tasks, MAI beat OpenAI's GPT-5.5 on quality at ten times better cost efficiency.
Why it matters
Microsoft owns the tools small teams already pay for: Copilot, VS Code, Microsoft 365. If its in-house models are genuinely cheaper to run, that pressure shows up in your bill, and it gives you a real alternative to being locked into one frontier lab. The catch is that a model tuned to run cheap on Azure is still lock-in, just a different vendor's.
Network impact
What to do
- If you use GitHub Copilot or VS Code, switch the model to MAI-Code-1-Flash on a real task and compare output quality side by side.
- Track your token spend before and after the swap. The whole pitch is cost, so measure it.
- Don't rip out OpenAI or Claude yet. Run MAI alongside for a week and keep what wins.
- If you build apps, request MAI-Thinking-1 access in Microsoft Foundry to test the reasoning model on your own data.
- Remember the lock-in risk: cheap models that only run well on Azure tie you to one cloud.