IT News on October 10th: On October 9th local time, Microsoft announced the launch of the Microsoft-Decision-1 decision-making model.

Officials state that this model is specifically designed for structured decision-making tasks and boasts top-notch performance in both response time and decision quality. Its actual performance surpasses that of large language models and other traditional decision-making models. This model has now been launched on Microsoft Foundry and has also been integrated with OpenRouter.

In a comparative evaluation that included 36 benchmark tests covering nearly 150,000 fully isolated training phases ( Blind Evaluation ), Microsoft-Decision-1 achieved the highest accuracy rate.

At the same time, benchmark test data shows that its response speed ranks first: it is 4.5 times faster than Quyet-1.0-Large, which is in second place, and 35 times faster than GPT-6 and Sol.

During the construction of Microsoft-Decision-1, Microsoft conducted post-training on Qwen3.5-9B to implement a single-inference decision scoring mechanism with ultra-low latency; subsequently, this will be used as a basis for migration and adaptation to other base models, including Microsoft AI ( MAI ) and OpenAI.

In addition, Microsoft conducted a system evaluation of Microsoft-Decision-1 based on 11 security benchmark tests, involving a total of 5,250 requests. The test scope covered harmful content, jailbreak attacks, and prompt injection. The results showed that while the model effectively intercepted various types of malicious activities, it still maintained a very high level of availability.

In terms of price, it costs $0.042 per million Token (Note from IT: the current exchange rate is approximately 0.28 RMB), and the output is free of charge.