What is Microsoft-Decision-1?

A decision model Microsoft released on October 9, 2026. Instead of generating text, it returns yes/no probabilities, choices or ordinal scores for text or JSON input, for classification, routing, ranking, evaluation scoring and choosing a model, tool or agent.

How is it different from a generative LLM?

Per Microsoft Learn, it doesn't produce free-form answers or written rationales. It makes one focused judgment about the input and returns a typed numeric decision. Use it when your app needs a predictable response shape and probabilities you can threshold or rank; use a generative LLM to create, summarize, rewrite or explain content.

Can it explain its reasoning?

No. Microsoft Learn says it returns numeric decisions without written rationales.

Can I use it on System One?

Not yet. It isn't enabled on System One; this page will be updated if that changes.

Where can I see pricing?

It isn't enabled on System One, so there's no credit price here. For Microsoft Foundry pricing, see Microsoft's official pages.

How does it relate to other decision models?

Microsoft Learn says its request format is similar to the `POST /v1/systemone` format: the app provides state and bounded questions, and the model returns decisions rather than generated text. For the concept and a comparison, see [What is a decision model?](https://system-one.dev/en/blog/what-is-a-decision-model)

Back to Blog

Microsoft-Decision-1: What It Does and Common Questions

by System OnePublished on Oct 11, 2026Updated Oct 11, 2026

Try it online (sign-in required): System One demos (these run models currently offered on system-one.dev, not Microsoft-Decision-1)

Use cases

Microsoft Learn lists five common uses (each with a related article on this site):

Gating and filtering: check whether content meets a condition before a workflow continues. See AI agent stop conditions and AI agent guardrails.
Classification and routing: assign tickets, alerts, documents or requests to one of a known set of options. See agent tool routing.
Ranking and prioritization: order incidents, leads, search results or review queues on an application-defined scale.
Evaluation and scoring: score model outputs against an ordinal rubric. See An LLM-as-a-judge alternative.
Choosing a model, tool or agent: pick the component that should handle a request. See LLM model routing.

To create, summarize, rewrite or explain content, Microsoft recommends a generative LLM instead. You can combine them: use Microsoft-Decision-1 to route or gate, then pass approved requests to a generative model.

Official sources

Microsoft Learn: Deploy and use Microsoft-Decision-1 in Microsoft Foundry: https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision
Microsoft launch blog: Introducing Microsoft-Decision-1 (2026-10-09): https://commandline.microsoft.com/microsoft-decision-1-model-foundry/

Checked: 2026-10-11

Frequently asked questions

What is Microsoft-Decision-1?: A decision model Microsoft released on October 9, 2026. Instead of generating text, it returns yes/no probabilities, choices or ordinal scores for text or JSON input, for classification, routing, ranking, evaluation scoring and choosing a model, tool or agent.
How is it different from a generative LLM?: Per Microsoft Learn, it doesn't produce free-form answers or written rationales. It makes one focused judgment about the input and returns a typed numeric decision. Use it when your app needs a predictable response shape and probabilities you can threshold or rank; use a generative LLM to create, summarize, rewrite or explain content.
Can it explain its reasoning?: No. Microsoft Learn says it returns numeric decisions without written rationales.
Can I use it on System One?: Not yet. It isn't enabled on System One; this page will be updated if that changes.
Where can I see pricing?: It isn't enabled on System One, so there's no credit price here. For Microsoft Foundry pricing, see Microsoft's official pages.
How does it relate to other decision models?: Microsoft Learn says its request format is similar to the `POST /v1/systemone` format: the app provides state and bounded questions, and the model returns decisions rather than generated text. For the concept and a comparison, see [What is a decision model?](https://system-one.dev/en/blog/what-is-a-decision-model)