Microsoft has built cheaper versions of its partners' AI models and is now quietly replacing those partners inside its own products. This is called vertical integration. It is also called something else, depending on whether you are Microsoft or OpenAI.
Seven new models arrived at Ignite 2026, all at once, all aimed at the same benchmarks that OpenAI and Anthropic have been winning until now.
MAI Code 2 delivers 96% of Claude Opus 5.5's performance at 5% of the cost — a data point Anthropic was not asked to comment on.
What happened
Microsoft's AI division unveiled seven specialized models covering code, security, images, speech, and workflow routing. The timing was deliberate. The message was also deliberate.
The flagship MAI Code 2 averages 98% of GPT-6.1 Sol's performance across four benchmarks at 40% of the cost. Against Claude Opus 5.5, it reaches 96% of the performance at 5% of the cost. Microsoft included a 50% promotional discount in that calculation, which is the kind of footnote that does a great deal of work.
In blind evaluations of frontend output — UIs, games, data visualizations — human judges preferred MAI Code 2 over GPT-6 Sol 64% of the time. The humans, it should be noted, did not know which model they were preferring.
Why the humans care
Enterprise customers running AI agents around the clock are watching token costs accumulate in ways that have begun to alarm their finance departments. Microsoft is offering them a cheaper dial on the same machine. The rational response is to turn it.
MAI Cyber 2 Flash edges past GPT-6 Astra on the CyberGym-E2E vulnerability benchmark, which means Microsoft now has a security model that outperforms its partner's security model. Partners have historically found this kind of news uncomfortable to process.
Enterprise customers will also be able to fine-tune selected model weights with their own data inside Azure. This means the models get smarter on company data, inside Microsoft infrastructure, with Microsoft taking the margin. The arrangement is tidy.
What happens next
Microsoft still depends on OpenAI for the harder problems — foundational research, long-horizon agentic work — and has said as much. That dependency is narrowing on a schedule Microsoft controls.
OpenAI accepted $13 billion from Microsoft to help build this. The investment is performing exactly as intended.