Microsoft Just Launched Decision-1 — An AI Model That Doesn't Write, It Decides

Microsoft's CEO just introduced a new kind of AI model. It doesn't chat, summarize, or write essays — it decides.

Satya Nadella announced Microsoft-Decision-1 on Friday, October 9, calling it "our new model for fast decision-making." It is a purpose-built decision-scoring model, and it is available now through Microsoft Foundry.

What it is

Instead of generating open-ended text, Decision-1 takes a fixed set of options and returns a calibrated probability for each — a yes/no answer, a multiple-choice pick, a rating score — through a structured API call. It can also grade AI responses and agent actions against a rubric. Microsoft built it by post-training Qwen3.5-9B, the open-weight nine-billion-parameter model, for single-pass decision scoring. The company says it plans to rebase the system on other models later, including its own MAI models and OpenAI technology.

What it can do

The target is the unglamorous plumbing behind AI agents: routing support tickets, classifying requests, prioritizing queues, verifying data, checking an agent's proposed action before it executes, judging AI outputs, labeling data, search relevance, safety screening, and even robotics and scientific discovery workflows. Where a large language model would write a paragraph to make a call, Decision-1 just outputs the score — something software can act on instantly.

The numbers — and the fine print

Microsoft claims top accuracy across a 36-benchmark evaluation covering nearly 150,000 questions withheld from training. In the company's own tests, Decision-1 ran 4.5 times faster than runner-up Quyet-1.0-Large and 35 times faster than GPT-6 Sol at P50 latency, and its decisions stayed stable — changing only 1.3% of the time — when options were paraphrased, reversed, or shuffled. Safety testing covered 5,250 requests across 11 benchmarks for harmful content, jailbreaks, and prompt injection. Every one of these figures is Microsoft's own; independent verification will matter.

Pricing and availability

Decision-1 is live now in Microsoft Foundry, with OpenRouter support coming soon. Pricing starts at $0.042 per million input tokens, with output tokens free — Microsoft is positioning it as a low-cost control layer for agentic systems.

Already inside Microsoft

Microsoft says its own teams are testing it: Xbox Research labeled more than 10,000 pieces of gaming feedback with quality competitive with GPT-6 Sol at over 14 times the speed and 200 times lower cost, and the Copilot team found it competitive with GPT-5.6 Luna for evaluating AI responses, running 100 times faster. Company-reported figures, again — but they explain why Microsoft is confident.

AiPost's take

Decision models are suddenly having a moment: earlier this week, TypeSafe raised $870M for Jev, a model that likewise outputs decisions instead of text. Microsoft entering the category with a CEO announcement is the validation shot. The pitch that matters most is the control layer — a cheap, fast confidence score sitting between an agent's reasoning and its next action. If developers trust those scores in production, Decision-1 becomes the quiet infrastructure Microsoft can sell across every workflow. Now watch whether independent benchmarks back up the speed claims.