Microsoft Decision-1: 9B Qwen Model Launch Explained

Microsoft Decision-1 is a specialized model for choosing among options rather than generating conversational text. Given input and a predefined set of answers, it returns a probability score for each in one pass, supporting tasks such as ticket routing, content classification, prioritization, ranking, and checking an AI agent’s output. The approach is intended to make high-volume decisions faster and cheaper, while avoiding the need to parse a chatbot’s prose or infer its confidence. Decision-1 is based on Alibaba’s Qwen3.5-9B, which Microsoft says is a temporary foundation; the company plans to rebase it on other models, including its own MAI models and OpenAI models.

Microsoft reports 83.5% accuracy across 36 benchmarks and nearly 150,000 held-out questions, along with speed advantages of up to 35 times over GPT-6 Sol and 98.7% stability under small input changes. These figures are company-reported and lack independent verification in the coverage described. The model is available through Microsoft Foundry and OpenRouter, but developers cannot download its weights to run it themselves. It accepts text inputs up to 32,000 tokens, does not provide explanations, and does not support images, audio, or video, so other models may be needed for those functions. OpenRouter lists a price of $0.042 per million input tokens with no charge for output, though Microsoft’s own materials cited do not confirm that rate.