Microsoft skipped OpenAI's decision model and built its own on Alibaba's Qwen (thenewstack.io)

Microsoft has launched its own decision model, Microsoft-Decision-1, leveraging Alibaba's Qwen3.5-9B, just days after OpenAI released its Decisions API for developers. This decision marks a strategic move away from OpenAI’s models, which Microsoft plans to incorporate into future iterations. With a cost of $0.042 per million input tokens, Microsoft aims to compete in a rapidly evolving landscape where other companies like TypeSafe and Cloudflare have already established decision models. Notably, Microsoft has begun internal testing of Decision-1 across various teams, showcasing its applications in managing player feedback for Xbox and enhancing responses in Microsoft’s Copilot tool. The significance of Decision-1 lies in its speed and consistency; it reportedly outperforms OpenAI's GPT-6 Sol by over 14 times in performance metrics for various applications. However, the model currently processes only text and returns JSON, distinguishing it from competitors that support images. While Microsoft claims that Decision-1 features calibrated probabilities, implying a predictive accuracy of about 90%, concerns remain about its robustness against adversarial inputs based on findings from other models like Jev. As Microsoft continues to refine its offering and address interoperability and calibration challenges, the launch could reshape decision-making tools within enterprise applications and influence the broader AI/ML community.