Microsoft has launched Microsoft-Decision-1.1, an enterprise-grade decision AI model now available across multiple platforms. Designed for high-frequency decision-making scenarios, the model handles structured judgment tasks such as routing and classification. Billing is based solely on input tokens, with no charge for output. The model makes choices within developer-defined rule constraints and can output calibrated confidence scores; when uncertainty is high, it can abstain or prompt for human intervention to reduce the risk of misoperation. Compared with the previous generation, its intelligence level and calibration capability have improved, its speed has more than doubled, and it outperforms multiple similar models in relevant benchmark tests. The model is optimized based on a specific medium-sized mixture-of-experts model and is compatible with a variety of GPUs. Its training combines supervised fine-tuning and reinforcement learning, and it supports multimodal input. Currently, more than 4,000 customers are using it, and it has been deployed in multiple scenarios within Microsoft.
