Microsoft launches Microsoft-Decision-1, a model built for fast decisions
Microsoft has introduced Microsoft-Decision-1, a model designed for fast decision-making on structured tasks. According to Microsoft CEO Satya Nadella, it outperforms both large language models and other decision models in latency and quality.
- Microsoft introduced Microsoft-Decision-1, a model for fast decision-making on structured tasks.
- According to Microsoft, it outperforms both LLMs and other decision models in latency and quality.
- It is being tested internally for incident response, quality control and scientific discovery.
- The model is available now in Foundry and is coming soon to OpenRouter.
- Figures on speed, price and architecture appear only in user comments, not in Microsoft's post.
A model for structured decisions
Nadella announced the model in a LinkedIn post, describing Microsoft-Decision-1 as "our new model for fast decision-making". The company says it delivers top performance on structured decision tasks. It also claims the model beats both general-purpose LLMs and other decision models on latency and on quality.
The post gives no benchmark tables or methodology. The performance claims are Microsoft's own and have not been independently confirmed in the source material.
Internal testing and availability
Microsoft says it is already testing the model across the company. The use cases named in the post are incident response, quality control and scientific discovery. These are areas where a system has to choose between a limited set of actions quickly, and where speed matters as much as accuracy.
Microsoft-Decision-1 is available now in Foundry, Microsoft's platform for building AI applications. According to Nadella, it will soon also be offered through OpenRouter. The post does not mention pricing, context limits or supported input types. One commenter asked whether image input will be supported, and the post does not answer that.
What commenters add, and what remains open
Replies under the post point to further details, but these come from commenters who say they read the linked article, not from the post itself. One commenter cited a speed of 35 times faster than GPT-6 Sol and a price of $0.042 per million input tokens. Another reported 85ms latency at 83% accuracy and said the model was post-trained from Qwen3.5-9B. A third commenter said the model reached the top 10 on a Pac-Man benchmark run by Opper AI. None of these figures appear in Nadella's own text, so readers should check them against Microsoft's documentation.
The discussion also shows what early adopters are asking about. Several commenters focused on calibration, meaning whether a model's confidence matches how often it is right. They also asked who sets the threshold at which a system acts on its own and when it hands a case to a human. Others said that a model's proposal and an action it is allowed to take should stay separate, with logging and human oversight in between, especially in incident response.
For developers, the practical point is that a dedicated decision model is now available in Foundry, with a second distribution channel promised. Whether it delivers the lead in latency and quality that Microsoft claims will depend on published evaluations and on how it performs in production workflows.
