Microsoft-Decision-1 scores fixed options with probabilities instead of text

Microsoft's Decision-1, post-trained from Qwen3.5-9B, returns a probability per answer option for routing and triage via Foundry and OpenRouter.

Microsoft has launched Microsoft-Decision-1, a model that returns a probability for each answer in a fixed list instead of writing text. Microsoft's launch post positions it for routing, classification, prioritization, verification and workflow control. If your agent pipeline uses a general LLM to answer "which team gets this ticket?" and then parses the reply, this is a cheaper, typed alternative. It is available in Microsoft Foundry and through OpenRouter.

Microsoft post-trained it from Qwen3.5-9B and plans to rebase it on MAI and OpenAI models, InfoWorld reports. Microsoft's Learn docs describe three question types: noul (yes/no probability), choice (one label plus a probability per option) and score (a position on an ordered scale). The quickstart needs an Azure subscription and Python 3.10 or later. OpenRouter lists the price at $0.042 per million input tokens, with output free.

The limits matter. The Foundry model card says it is text-only, takes up to 32K tokens, gives no rationale, and should not be the sole decision-maker in consequential decisions about people. The accuracy and speed claims are Microsoft's own: the highest accuracy across 36 benchmarks, and 35 times faster than GPT-6 Sol. InfoWorld notes these have not been independently verified. The New Stack reports that Microsoft has not announced open weights. It also reports that the JevOut adversarial study did not test Decision-1, so robustness against misleading inputs is still unknown.

It also faces competition. The New Stack notes that Jared Palmer lists his open-source Kev-4B on OpenRouter at the same input price. Microsoft's docs advise checking calibration and thresholds against your own labeled examples before you automate a gate.

Sources