Microsoft has introduced Microsoft-Decision-1, a specialized AI model designed to select from fixed answer options rather than generate text. This model is built upon Alibaba’s open-weight Qwen3.5-9B and enhanced through Microsoft’s post-training, delivering high accuracy and rapid processing speeds in decision-making tasks.
- Built on Alibaba’s open-weight Qwen3.5-9B model
- Achieved highest accuracy in 36 benchmark tests
- Up to 35 times faster than OpenAI’s GPT-6 Sol in latency
What happened
Microsoft has launched Microsoft-Decision-1, a compact AI model that selects from fixed answer options instead of generating free-form text. The model was constructed using Alibaba’s open-weight Qwen3.5-9B and subsequently post-trained by Microsoft to optimize its decision-making capabilities. The launch was announced by Achint Srivastava, a vice president in Microsoft's Office of the CTO, highlighting the emergence of decision models as a significant AI category.
The model supports various types of decisions including yes/no, multiple choice, and ratings, returning calibrated probability scores for each possible answer. Its large 32,000-token context window allows it to analyze extensive text inputs. Microsoft reports that Microsoft-Decision-1 scored the highest accuracy across a series of 36 benchmarks involving nearly 150,000 questions kept blind from training.
Why it matters
Decision models like Microsoft-Decision-1 represent a distinct class of AI focused on evaluation and selection rather than text generation. This specialization allows for faster, more efficient processing ideal for practical tasks such as routing, classification, prioritization, and workflow control. Microsoft-Decision-1’s speed and accuracy make it attractive for enterprise applications requiring reliable, discrete decision outputs.
Compared to leading models like OpenAI’s GPT-6 Sol and H2O-Lightning-4B, Microsoft-Decision-1 performed up to 35 times faster with competitive or superior accuracy. Its cost effectiveness is also notable, with input costing $0.042 per million tokens and no charge for outputs. This combination of performance and efficiency positions the model well in a growing market of AI decision tools supported by major providers including Cloudflare and Snowflake.
What to watch next
Microsoft plans to further develop Microsoft-Decision-1 by rebasing it on other models, including its own MAI series and OpenAI’s offerings. This will likely refine its capabilities and integration potential within Microsoft’s broader AI ecosystem. Observers should monitor how this model is adopted in areas such as AI response grading, verification, and automated workflow management.
The competitive landscape for decision models is rapidly expanding, with more than 100 similar models now available, including TypeSafe AI’s Jev and offerings from startups and large tech firms alike. Microsoft-Decision-1’s combination of speed, accuracy, and cost efficiency sets a benchmark, but ongoing innovation and benchmarking against emerging models will determine its long-term impact.