Microsoft releases Decision-1, a small model that picks between fixed options instead of writing text

What launched

Microsoft released Microsoft-Decision-1 on 9 October 2026.¹ It is a decision-scoring model: you give it a question and a fixed set of options, and it returns a calibrated probability for each option in a single structured API call, instead of generating open-ended text.¹²

Microsoft says it is meant for routing, classification, prioritisation, verification and workflow control, including grading AI responses and agent actions against a rubric.¹ It supports yes/no, multiple-choice and rating formats.¹²

How it is built

The model is post-trained from Alibaba’s open-weight Qwen3.5-9B for single-pass scoring.¹² Microsoft says it will soon rebase the model on other models, including Microsoft AI and OpenAI models.¹

Price and availability

Microsoft lists $0.042 per million input tokens, with output tokens free.¹² It is available in Microsoft Foundry, and Microsoft’s post also lists OpenRouter.¹ TestingCatalog described OpenRouter support as coming soon at the time it reported.²

Microsoft’s claims

Microsoft says the model had the highest accuracy in a 36-benchmark comparison of nearly 150,000 questions held out from training, and that its median response time is about 35 times faster than GPT-6 Sol.¹ It also says decisions changed on 1.3% of reworded inputs on average.¹ These are Microsoft’s own results and have not been checked independently.

Microsoft also cites internal teams. For example, it says Xbox Research found quality competitive with GPT-6 Sol at over 14 times the speed and 200 times lower cost.¹

Who made it

RuntimeWire reports that the post is by Achint Srivastava, a Microsoft vice president who earlier co-founded the AI evaluation start-up Pi Labs.³

What is not stated

Microsoft’s post does not state a licence for the model’s weights.¹