AIAIAI.FYI
LIVE

Microsoft AI releases seven MAI models at Ignite, pitched as cheaper than OpenAI and Anthropic

Microsoft says MAI Code 2 reaches 98 percent of GPT-6.1 Sol's benchmark performance at 40 percent of the cost, using a 50 percent promotional price.

What happened

Microsoft has built its own AI models for coding, security, images, speech and routing. It unveiled seven of them at its Ignite conference. It says they do similar work to models from OpenAI and Anthropic at lower prices.

Why it matters to you

Companies that run AI agents around the clock pay for every token, the units AI models read and write. Cheaper models can cut that bill.

The takeaway

Microsoft is pushing its own cheaper AI models in place of outside ones, though the performance claims come from Microsoft's own tests.

What it means for you

If you build software

MAI Realtime 1 is compatible with the OpenAI Realtime API, so existing voice apps can switch over easily, and MAI Code 2 is available in GitHub Copilot.

If you run a business

Microsoft lists MAI Cyber 2 Flash at $0.10 per million input tokens versus $10 for GPT-6 Astra, and companies can ask their Microsoft account team about MAI Licensed Weights.

If you invest

Microsoft is moving its own products to in-house models while keeping a non-exclusive license to OpenAI's models through 2032, after the April restructuring of the partnership.

Get stories like this every morning