Science & Tech · AI · 1 day ago
Microsoft launches decision model built on Alibaba’s Qwen AI
Microsoft has introduced an AI model designed to classify information and make set decisions rather than write open-ended text.
Its first version is built on a model from Alibaba Cloud.
Microsoft says it plans to move future versions to its own models and those from OpenAI.
The company says the model is fast and inexpensive, but decision models can still make mistakes.
Microsoft launched Microsoft-Decision-1, a decision model offered through Microsoft Foundry and expected soon on OpenRouter.
The model’s first version is based on Qwen3.5-9B, developed by Alibaba Cloud.
Microsoft said a future version will be based on its own models and models from OpenAI, without giving a reason for the change.
The company reported that Microsoft-Decision-1 achieved 83.5 percent accuracy across 36 benchmarks and a confidence score of 92.2 percent.
Microsoft said the model is 2.5 times faster than H2O-Lightning-4B and 2.8 times faster than Jev in its latency test.
Input tokens cost $0.042 per million, while output tokens are free.
- Who
- Microsoft
- What
- Launched Microsoft-Decision-1, a decision model for structured outputs and classification tasks.
- When
- October 2026; the article was published on October 10.
- Where
- Microsoft Foundry; OpenRouter availability is planned.
- Why
- Microsoft said decision models can deliver structured outputs that software can act on, and that cost matters when choosing AI models.
This story does not have two clearly opposing sides.
Decision models are quickly emerging as an important new category in AI
And it’s increasingly important to choose the right model for the right job.
This story does not have a timeline yet.
- Model
- Microsoft-Decision-1
- Initial base model
- Qwen3.5-9B, from Alibaba Cloud’s Qwen model family
- Reported accuracy
- 83.5 percent across 36 benchmarks
- Reported confidence score
- 92.2 percent; second behind Quyet-1.0-Large
- Input price
- $0.042 per million tokens; output tokens are free











