- Microsoft-Decision-1 uses Alibaba’s open-source Qwen3.5-9B model and delivers faster, lower-cost scoring for AI applications
- Microsoft is testing the technology across areas including incident response, quality control and scientific discovery
Microsoft has launched a new AI decision model built on Alibaba’s open-source Qwen3.5-9B, highlighting Qwen’s growing role as a foundation for specialized AI systems.
The model, called Microsoft-Decision-1, was released on October 9 and is available through Microsoft Foundry and OpenRouter. Rather than being trained from scratch, it was developed by further training Qwen3.5-9B, according to Microsoft.
AI that scores, not writes
Unlike conventional large language models that generate text or reason through problems in natural language, decision models return structured probability scores for predefined choices.
Given a yes-or-no question, a multiple-choice task or a rating request, the model assigns confidence scores to the available options. Software can then use those scores to decide whether to proceed, wait or escalate a case to a human.
“Decision models are rapidly emerging as an important new category of AI,” wrote Achint Srivastava, corporate vice president of software engineering in Microsoft’s Office of the CTO, in the launch announcement.
Faster and cheaper
Microsoft said Decision-1 achieved the highest accuracy across 36 internal benchmarks, with a median latency of 85 milliseconds — 35 times faster than GPT-6 Sol.
The Xbox research team used the model to classify more than 10,000 pieces of player feedback, achieving quality comparable to GPT-6 Sol at more than 14 times the speed and about one-200th of the cost. The Copilot team used it to evaluate answer quality, with speed improvements of up to 100 times.
Microsoft lists pricing at $0.042 per million input tokens, with output tokens free.
Microsoft CEO Satya Nadella said the company was testing the model across a range of fields, including incident response, quality control and scientific discovery.
Qwen’s expanding role
Microsoft’s choice of Qwen reflects a wider pattern in the emerging decision-model market. Of 129 open-source decision-model projects on Hugging Face, 84 are based on Alibaba’s Qwen models, accounting for more than 65%, according to the figures cited in the report.
Other companies using Qwen as a foundation for decision models include Perplexity, Cloudflare and H2O.ai.
Microsoft said it plans to develop related versions based on its own MAI models and OpenAI models. The initial release, however, shows how an open-source model developed by a Chinese company can serve as a starting point for specialized AI systems built by global technology firms.
As decision models move into production, their appeal will depend not just on benchmark scores, but on whether they can make reliable, low-latency decisions at a cost that makes large-scale deployment practical.
Header image generated by Tencent’s Yuanbao AI
