AI news · Microsoft · models
Microsoft launches Decision-1, a fast model for software choices
Built on Alibaba’s Qwen3.5-9B, it scores fixed options with a probability and joins more than 100 rivals to TypeSafe’s Jev.
Microsoft released Microsoft-Decision-1 on Friday 9 October 2026, a model that picks among fixed options and returns a calibrated probability for each, the company says. It is built for routing, classification, prioritisation and checking agent actions, and is offered in Microsoft Foundry and through OpenRouter.
The first version is Alibaba’s open Qwen3.5-9B, post-trained by Microsoft, which says it plans to rebase the model on its own MAI models and on OpenAI’s. Microsoft ran 36 benchmarks with nearly 150,000 questions and puts Decision-1 top on accuracy. On speed, it beats the runner-up, H2O-Lightning-4B, by 2.5 times and GPT-6 Sol by 35 times. Those are Microsoft’s own figures; Qwen comes from Alibaba, The Register notes.
Input costs $0.042 per million tokens and output is free, The Register reports. It counts more than 100 decision models now chasing TypeSafe’s Jev, launched three weeks earlier, including entries from OpenAI, Cloudflare, Perplexity and Snowflake.
For teams running agents and ad workflows, routing and scoring steps now come from the largest software vendors at a fraction of a chat model’s price.
Sources
- Microsoft: Introducing Microsoft-Decision-1, our model for fast decision-making. Checked
- The Register: Microsoft leans on open weight model from Chinese AI lab to challenge Jev. Checked