Microsoft introduces new decision-making model

Microsoft has launched Microsoft-Decision-1, a decision-making model aimed at reducing the cost of small arbitration tasks in applications. The model is based on Qwen3.5-9B, a 9-billion-parameter model developed by Alibaba, and is available in public preview on Microsoft Foundry and OpenRouter.

Model helps route requests

The model assigns probabilities to different options for handling a request, such as directing a customer service query to billing, technical support, or sales. It can then route the request or ask for human help if the result is unclear. This can be useful for software publishers, IT teams, and companies building AI agents.

Limitations and usage guidelines

The model works only with text up to about 32,000 tokens and does not explain its decisions. Developers must set possible answers, confidence thresholds, and situations requiring human input. Microsoft warns against using it as the only tool for decisions like granting credit or hiring people.

Cost and efficiency

Microsoft charges 0.042 dollars per million input tokens, with no charge for outputs. This rate targets applications that handle sorting and control operations, often at each step of an AI agent’s work. In one example, Xbox Research used the model to categorize over 10,000 player feedbacks efficiently.

Competitor landscape

Microsoft is entering a competitive market. Jev from TypeSafe AI offers a similar service with the same pricing. OpenAI provides a public beta API that accepts both text and images, expanding its use cases. Other companies offer models like H2O-Lightning-4B, which process text and images but require more technical setup.