Microsoft has launched a specialized AI model, Microsoft Decision-1, designed to make specific decisions rather than generate free-form text. The goal is to offload routine judgment tasks away from large language models (LLMs) and into a leaner model that Microsoft says runs faster and cheaper.
Instead of writing paragraphs, Decision-1 returns a probability for each possible answer, letting software applications consume its output directly. The model targets tasks such as classifying requests, prioritizing incidents, and checking AI-generated responses.
What Microsoft Decision-1 actually does
According to Microsoft’s blog post, “Instead of writing a free-form response, it takes a fixed set of answer options and returns a calibrated probability for each one in a single pass.”
So where a general-purpose LLM might generate a paragraph of explanation, Decision-1 outputs numbers. It supports yes-or-no judgments, choices among predefined options, ratings, and rubric-based assessments of AI responses and agent actions.
Developers can feed those outputs into software to classify inputs, rank work, verify responses, or pick the next step in an AI agent workflow. Microsoft says the model can also evaluate multiple decisions in parallel within a single request.
This fits a broader trend vendors are pursuing as AI agents grow: carving out a separate “decision-making” layer that sits in front of — or replaces — a big generative model, according to reporting from InfoWorld.
Benchmarks and pricing
Microsoft claims Decision-1 topped a comparison of 36 benchmarks covering nearly 150,000 questions that were kept blind from training. The company says the model is about 2.5 times faster than the runner-up and 35 times faster than GPT-6 Sol.
On pricing, Microsoft lists Decision-1 at 4.2 cents per million input tokens, with no charge for output tokens.
To put that in perspective, the company worked through an incident-prioritization example: a request using 86 input tokens and two output tokens cost approximately $0.0000036. At that rate, $1 would cover more than 275,000 similar decisions.
Microsoft noted these benchmark results and cost calculations are its own claims and have not been independently verified.
Governance and accuracy concerns
The model returns scores without explaining why. Microsoft’s documentation also warns that results can shift depending on how questions are worded.
Charlie Dai, vice president and principal analyst at Forrester, said specialized decision models could cut the cost and latency of agentic AI workflows. “Routing, classification, and approval checks often represent a large share of agent interactions, so a specialized decision model can significantly reduce cost and latency while reserving general-purpose LLMs for reasoning and content generation,” Dai said.
But he also flagged a governance gap. “The absence of reasoning can make decisions harder to explain, validate, and challenge,” Dai said, recommending calibration testing, controls over question wording and score thresholds, audit logging, and human review for high-impact decisions — especially when customer outcomes or regulatory obligations are involved.
Amit Kumar Jena, development manager for AI at Kanerika, called moving routine routing and approval checks away from large generative models a meaningful architectural change for enterprises scaling AI agents. But he cautioned against using Microsoft’s benchmark numbers to predict real savings. Organizations should measure “cost and latency per successfully completed workflow, including fallbacks and mistakes,” he said.
Jena’s concern runs deeper: “My concern is that a precise-looking probability can give an uncertain decision more authority than it deserves.” His advice is to start with reversible tasks like request routing and triage, and require meaningful human review for decisions that materially affect people.
What this means for you
For most Windows users, Microsoft-Decision-1 won’t show up on your desktop anytime soon. It’s a backend model aimed at developers and enterprises building AI agents. But the idea behind it — splitting cheap, fast decisions from expensive, creative generation — could shape how AI tools behave behind the scenes.
If you use AI-powered apps that route tickets, flag content, or triage requests, this kind of model could make them faster and cheaper to run. The catch is transparency: when a decision affects you, you may want to know whether a human was in the loop.
How to get Decision-1
Microsoft-Decision-1 is available through Microsoft Foundry and OpenRouter. The company built the model by post-training Qwen3.5-9B and says it plans to rebase it on other models, including Microsoft AI (MAI) and OpenAI models.
For developers, the practical move is to test it on a single, well-defined decision task and measure real cost and latency before rolling it out — the same caution analysts are raising.
Source: Computerworld
Over to you: Would you trust a probability score to make a decision that affects you, or do you want a human in the loop every time?
