Microsoft-Decision-1

Microsoft-Decision-1 is a small decision model from Microsoft, post-trained from Qwen3.5-9B and offered as a hosted model in Microsoft Foundry. Instead of writing free text, it reads a prompt together with a fixed set of answer options and returns a calibrated probability for each option in a single forward pass. It handles yes or no questions, multiple choice and ratings, and it can grade AI answers or agent actions against a rubric. Microsoft aims it at routing, classification, prioritisation, verification and workflow control, where a fast and consistent score matters more than prose. Its announcement gives an input price of $0.042 per million tokens with no charge for output, and says the model is also coming to OpenRouter. The announcement does not state a parameter count, context length or licence, and links no downloadable weights, so for now it is a hosted service only. Microsoft reports its accuracy on a comparison of 36 benchmarks and its speed against other models, but these are the vendor's own tests.

Not yet independently verified: Read on Microsoft's own announcement of 9 October 2026 and its Foundry catalogue page; no independent report or measurement was found. The accuracy and speed comparisons are Microsoft's own, so no score is stored, and the announcement gives no parameter count, context length or licence. We will update this when it can be confirmed, and remove this note.