The model is in public preview. Its weights are not released. It handles yes/no, multiple-choice and rating questions, plus rubric-based grading of AI responses and agent actions. In Microsoft's own tests across 36 benchmarks, it hits 83.5 percent accuracy with 85 ms latency. It also ran 2.5 times quicker than H2O-Lightning-4B v1.1, the runner-up. The Copilot team found it competitive with GPT-5.6 Luna and 100 times faster. No outside group has checked these figures yet. Microsoft is arriving late to this category. Similar models from OpenAI, Cloudflare, and the "original" from Jev came first, and Microsoft's pricing aligns with TypeSafe AI's Jev. Vercel also offers it through its AI Gateway, and OpenRouter access is planned. Microsoft says later versions will be rebuilt on top of OpenAI and its own MAI models. For founders, choosing between options inside an agent is now a cheap, swappable building block, and Microsoft's distribution makes it hard to compete on price alone. Startups selling routing or evaluation layers will need proof on their own customers' data, because every published benchmark so far comes from the vendors themselves.