Microsoft launches Decision-1, a hosted model for scoring options, built on Alibaba's open Qwen3.5-9B

Microsoft

Microsoft has launched Microsoft-Decision-1, a model built to score a fixed set of options rather than write long text. Given a question, context and candidate answers, it returns probabilities in a structured form that software can act on, for tasks such as classification, ranking and routing. Microsoft further trained Alibaba's open-weight Qwen3.5-9B for the job, and the model is offered as a hosted service with its weights not distributed.

• Available in public preview through Microsoft Foundry, with OpenRouter named as another access route; input tokens cost $0.042 per million and output tokens carry no separate charge.
• Microsoft says it is 2.5 times faster than H2O-Lightning-4B in median latency and has the highest accuracy across 36 benchmarks; these are its own internal results.
• The model is built for structured decisions and is not shown to beat general-purpose models on writing, research, coding or complex reasoning.
• Microsoft says future versions may use other foundations, including models from Microsoft AI and OpenAI.
• The article notes that a model judging another model's output needs testing, and that the cost per decision depends on request size.