AI Tech Flash ── 2026-10-17
Microsoft releases a decision-focused model
A model that picks among set options enters preview in Foundry
What changes
Developers can hand classification and routing steps to a dedicated model instead of relying on a text-generating one. In Microsoft's announcement, it is meant to work alongside generative and reasoning models.
One example Microsoft gives is a language model drafting a reply while Microsoft-Decision-1 checks that reply against defined criteria.
Why it matters
According to Microsoft, an AI application may make thousands of small choices around a comparatively small number of complex generative or reasoning tasks. The company says running every step through the same general-purpose model can overlook an opportunity to optimize.
In Microsoft's announcement, decision models are aimed at these steps, with lower latency and lower cost as the goal.
What it is
Microsoft-Decision-1 is a decision model that chooses among predefined options. According to Microsoft, it is built on Qwen3.5-9B. It is now in public preview.
It does not generate free-form text. According to Microsoft Learn, it reads text or JSON and returns a typed, numerical decision. It gives no written rationale.
Who and where
Microsoft built it, and it is available in the Microsoft Foundry model catalog and the Foundry Playground. You deploy it through the Foundry portal or the Azure CLI.
According to Microsoft Learn, deployment types are DataZoneStandard in selected regions and GlobalStandard. An Azure subscription is required.
The New Stack reports that Microsoft launched it on a Friday. The outlet says this was three days after OpenAI opened its Decisions API to every developer in public beta.
Why it was built
According to Microsoft, many of the most frequent operations in AI applications are decisions about what should happen next, not generation problems. Examples include classifying an input, selecting an option, or verifying a condition.
The company says such simple, repeatable decisions do not need the generative capabilities or extended reasoning of a larger language model.
How it works
An application sends a state, which can be text or JSON, and typed questions. According to Microsoft Learn, there are three question types.
A noul question asks true or false and returns a probability from 0 to 1. A choice question selects one option and returns a probability for each option. A score question rates an item on an ordered scale and returns a probability-weighted average of the levels.
The documentation advises escalating to a person or another model when the top probability is low. It also says to validate thresholds on your own labeled examples before automating.
Key points
- Microsoft says the model is built on Qwen3.5-9B and is in public preview.
- It does not generate text. It returns typed decisions with probabilities.
- Microsoft says Xbox Research reported quality competitive with GPT-5 when categorizing more than 10,000 pieces of feedback.
Conclusion
Microsoft-Decision-1 is a model specialized for decisions such as classification and routing. Microsoft's announcement envisions using it alongside generative and reasoning models.
Benefits in performance and cost must be checked on your own data. Microsoft itself asks users to validate thresholds and scores on their own examples.
Quotes and sources
“Now available in public preview, Microsoft-Decision-1 is a decision model built on Qwen3.5-9B and designed for applications that need to make fast, structured choices among predefined options.”── source
“Unlike a generative large language model (LLM), Microsoft-Decision-1 doesn't generate a free-form response or written rationale.”
“Microsoft on Friday launched Microsoft-Decision-1 in Foundry, three days after OpenAI opened its Decisions API to every developer in public beta.”
Source articles
Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.