Microsoft used its Ignite conference this week to unveil seven new AI models of its own — three for coding, two for cybersecurity, one for voice and one for decisions — and to place them inside GitHub Copilot, Microsoft 365 Copilot and Azure AI Foundry.
The headline claim is price. Microsoft says its MAI-Code-2 model delivers 98 percent of the performance of OpenAI's GPT-6.1 Sol at 40 percent of the cost, and 96 percent of Anthropic's Claude Opus 5.5 at just 5 percent of the price — figures from Microsoft's own benchmarks, some at promotional pricing.
What was announced
The coding trio — MAI-Code-2, MAI-Code-2-Flash and MAI-Code-2-Max — is generally available now in GitHub Copilot and Azure AI Foundry. MAI-Cyber-2-Flash scored 92.3 percent on the CyberGym-E2E security benchmark, slightly ahead of OpenAI's GPT-6 Astra, at what Microsoft says is less than 2 percent of the cost.
“Our goal is to provide the absolute best price performance available in the market today.”— Microsoft, announcing the MAI model family at Ignite
A voice model, MAI-Voice-2, and a decision model, MAI-Decision-1.1, round out the family. Microsoft also added models from other labs — including OpenAI's GPT-6.2 and Meta models — to Foundry, and announced Copilot Workspaces, a shared space where teams and AI agents work on projects together.
A caveat in the fine print
The performance claims come with an asterisk the company disclosed itself. Microsoft said it removed 92 of 905 tasks from one major benchmark after finding the tasks overlapped with its own training data — an unusually public admission in an industry where benchmark scores are marketing currency.
CEO Satya Nadella framed the moment in his Ignite keynote as an industry shift from scaling intelligence to scaling utility — getting more useful work out of every model, at lower cost.
What Happens Next
The models are live in Copilot now, so the test is whether developers and companies switch from the OpenAI and Anthropic models they already use. Watch for independent benchmark results — and for how OpenAI and Anthropic respond on price.



Loading comments…