AnswersPublished17 sources
Microsoft-Decision-1: How Microsoft’s AI Decision Model Works
Announced October 9, 2026, Microsoft Decision 1 scores fixed options with probabilities instead of generating prose; input pricing is $0.042 per million tokens, with output free. The model is built for structured tasks such as classification, routing, and evaluation, with reported formats including yes/no, multiple...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Announced October 9, 2026, Microsoft Decision 1 scores fixed options with probabilities instead of generating prose; input pricing is $0.042 per million tokens, with output free.
- The model is built for structured tasks such as classification, routing, and evaluation, with reported formats including yes/no, multiple choice, ratings, and rubric based grading.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What is Microsoft-Decision-1, launched by Microsoft on October 9, 2026, and how does it select fixed options with calibrated probabilities r. Article summary: Microsoft-Decision-1 is a specialized model announced on October 9, 2026, for choosing among predefined answers rather than writing open-ended responses. Its intended role is a fast, inexpensive decision layer for classi. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
Microsoft-Decision-1 is a specialized model for tasks where software needs to choose among predefined answers—not generate an open-ended response. Microsoft announced it on October 9, 2026, positioning it for structured decisions such as classification, routing, and evaluation.1 The distinction matters for AI agents and workflows: a system can use a choice score to decide whether to route, retry, or hand off a task, rather than asking a general-purpose model to write an explanation at every step.
How Microsoft-Decision-1 scores choices
An application provides context and a fixed set of possible answers. Microsoft-Decision-1 returns a probability for each option in a single scoring pass, rather than composing a free-form response.14 The probabilities are meant to express the model’s confidence across the supplied choices; they are not guarantees that a selected answer is correct.
Reported task formats include yes/no questions, multiple choice, ratings, classification, and rubric-based grading.6
22 These describe the kinds of decisions the model is intended to support; developers should check the model’s current documentation for the exact input and output format required by their application.
For example, a support workflow could give the model several routing categories and use the scores to select a destination. An agent could also use scores to choose between acting, retrying, or requesting human review. Those are possible workflow designs, not evidence of a specific measured end-to-end speed improvement.
Why use a decision model instead of a chat model?
Many software steps need a structured choice, not a paragraph. A model designed to score a limited set of options can return a result that an application can use directly. Microsoft describes Decision-1 for decision and classification workloads, including routing and evaluation.1
That focus also defines its limits: it is not a substitute for a model that must explain an answer, draft content, or handle an open-ended conversation. In practice, a team might use a text-generating model for content and a decision model for a narrow task such as selecting a category or evaluating candidate responses.
Model foundation, availability, and price
Microsoft says Decision-1 is built on Alibaba’s Qwen3.5-9B and further trained for decision scoring.1 The company’s launch post lists the model in Microsoft Foundry’s catalog as a public preview.
1
Published pricing is $0.042 per million input tokens, with no charge for output tokens.11 Whether that cost is attractive depends on the workload, including how much context the application sends and how often it needs a decision.
What Microsoft’s performance claims show—and don’t show
Microsoft says Decision-1 ranked first in its comparison across 36 benchmarks covering nearly 150,000 questions. The company also reports latency results of 4.5 times faster than Quyet-1.0-Large and 35 times faster than GPT-6 Sol.12
5
These are reported benchmark results, not independent confirmation that Decision-1 will be faster or more accurate for every application. Benchmark performance can depend on the tasks, comparison models, and test setup. Teams should evaluate the model on their own data and measure decision quality, latency, and the consequences of incorrect choices before relying on it in a workflow.
The practical idea is straightforward: use a model built to score constrained choices when the job is making a structured decision. Its value will depend on whether that focused approach works reliably for the particular decisions an application needs to make.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Microsoft-Decision-1: How Microsoft’s AI Decision Model Works"?
Announced October 9, 2026, Microsoft Decision 1 scores fixed options with probabilities instead of generating prose; input pricing is $0.042 per million tokens, with output free.
What are the key points to validate first?
Announced October 9, 2026, Microsoft Decision 1 scores fixed options with probabilities instead of generating prose; input pricing is $0.042 per million tokens, with output free. The model is built for structured tasks such as classification, routing, and evaluation, with reported formats including yes/no, multiple choice, ratings, and rubric based grading.