Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and

Microsoft has introduced Microsoft-Decision-1, a model optimized for structured decision tasks, claiming superior performance compared to large language models and other decision models in both latency and decision quality. The model is currently being tested internally at Microsoft for applications such as incident response, quality control, and scientific discovery. It is available immediately in Microsoft Foundry and will soon be accessible via OpenRouter.

Microsoft-Decision-1 delivers a 90% calibration rate, meaning predictions at this confidence level are correct approximately nine out of ten times, enabling applications to act, defer, or request review based on reliability. The model operates at 35 times the speed of GPT-6 Sol and costs $0.042 per million input tokens, positioning it as a cost-effective solution for high-volume decision-making in enterprise workflows.

The model’s design emphasizes structured decision thresholds, where organizations must define clear rules for when to act autonomously versus when to escalate to human review. This approach is critical in workflows like incident response or quality control, where incorrect decisions can have rapid downstream effects. Microsoft highlights the need for governance, fallback mechanisms, and transparency in decision-making processes.

Microsoft-Decision-1 has been benchmarked on the JEVman platform, achieving a top-10 ranking in AI decision-model performance, including tasks modeled after Pac-Man gameplay. The model is built on a fine-tuned version of Qwen3.5-9B, with internal testing showing a 46-fold improvement in consistency over prior state-of-the-art LLMs for adaptive replanning in scientific discovery workflows.