Microsoft launches decision model built on Alibaba’s Qwen3.5-9B
Microsoft releases Microsoft-Decision-1, a specialized AI model that selects fixed options rather than generating text, based on Alibaba’s open-weight foundation.
Key points
- Microsoft-Decision-1 is built on Alibaba’s Qwen3.5-9B and post-trained for classification tasks.
- The model claims to be 35 times faster than OpenAI’s GPT-6 Sol in internal benchmarks.
- Input costs are $0.042 per million tokens, with output tokens provided free of charge.
Microsoft announced Microsoft-Decision-1 on 9 October, a specialized artificial intelligence model designed to select from fixed answer options rather than generate open-ended text. Achint Srivastava, vice president of software engineering in Microsoft’s Office of the CTO, introduced the tool in a blog post that outlined its technical foundations and performance metrics.
The model is built upon Alibaba’s open-weight Qwen3.5-9B foundation, according to the Microsoft Foundry model card. Microsoft post-trained this base to create a system that returns calibrated probability scores for specific choices. The resulting tool handles yes-or-no questions, multiple-choice formats, and rating scales, allowing it to grade AI responses against predefined rubrics.
Technical specifications and performance
Microsoft-Decision-1 accepts text-only inputs and features a 32,000-token context window. In Microsoft’s internal testing, the model achieved the highest accuracy across a 36-benchmark comparison that included nearly 150,000 questions kept blind from the training data. The company stated that the model changed its decision only 1.3% of the time when requests were reworded or reordered.
Performance tests indicated Microsoft-Decision-1 was 2.5 times faster than the runner-up, H2O-Lightning-4B v1.1. It also performed 35 times faster than OpenAI’s GPT-6 Sol. Microsoft later updated its announcement to include benchmarks against Jev, a model from TypeSafe AI launched three weeks prior. The Register reported that Microsoft found its new model 2.8 times faster than Jev in latency tests.
Use cases and safety considerations
Microsoft lists routing, classification, prioritization, verification, and workflow control as primary use cases for the model. Xbox Research utilized Microsoft-Decision-1 to sort more than 10,000 pieces of player feedback and reviews into thematic categories. The company stated this process was competitive with GPT-6 Sol on quality, over 14 times faster, and 200 times cheaper.
The Copilot team found the model competitive with GPT5.6 Luna and 100 times faster in their evaluations. Microsoft tested the system on 5,250 requests across 11 safety benchmarks covering harmful content, jailbreak attempts, and prompt injection. The model card explicitly advises that the tool should not be the sole basis for consequential decisions about people, such as credit or employment matters.
What should you do?
Organizations evaluating specialized AI models should review the model card for explicit limitations regarding high-stakes decision-making. Security teams should monitor for prompt injection vulnerabilities, even in classification models, by testing against the safety benchmarks Microsoft cited. Developers should verify the cost structure, noting that input tokens incur charges while output tokens are free.
Background: Prompt injection attacks
Look for sudden shifts in tone, refusal to follow system constraints, or output that repeats hidden instructions. These signs indicate the model is processing injected commands rather than your intended query.
Read the full guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
What to do and how to stay safe: Microsoft
- Review the official model card for Microsoft-Decision-1 to understand its context window limits and safety guidelines.
- Test the model against your specific classification rubrics to validate accuracy before integrating it into production workflows.
- Monitor for updates as Microsoft plans to rebase the model on other architectures, including OpenAI and MAI models.
- Assess whether the speed and cost benefits align with your organization’s need for high-volume text routing and verification.
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the cost of using Microsoft-Decision-1?
Input tokens cost $0.042 per million tokens, while output tokens are free.
What foundation model is Microsoft-Decision-1 based on?
It is built on Alibaba’s open-weight Qwen3.5-9B model, which Microsoft post-trained.
How does Microsoft-Decision-1 compare to GPT-6 Sol in speed?
Microsoft states the model is 35 times faster than OpenAI’s GPT-6 Sol in its internal tests.
