On October 12th, Microsoft launched the new decision model Microsoft-Decision-1 based on Alibaba Qwen. According to Microsoft’s official announcement, this model was trained using Qwen3.5-9B. In structured decision-making tasks, the model outperforms popular decision models such as LLM and Jev 1.13.0 in terms of latency and accuracy. The Microsoft Xbox research team has used it to process over 10,000 player feedback items. Microsoft CEO Nadera also revealed on social media that, due to the excellent performance of the new model, Microsoft is testing the application of the new model in all areas, including event response, quality control, and scientific discovery.
It is understood that the Jev model’s popularity in September triggered a wave of decision-making models. These models do not generate natural language, but output predictions, probabilities, and confidence levels. They are much smaller in size than large language models LLM, and respond more quickly. This precisely meets the high demand for frequent judgments in AI tasks. Statistics show that among the 129 open-source decision model projects on HuggingFace, 84 are based on Alibaba Qwen, accounting for over 65%. Well-known decision models launched by companies such as Microsoft, Perplexity, Cloudflare, and H2O.ai all use Alibaba Qwen.