Microsoft launches Decision-1 AI model for fast decisions

Microsoft ↗ 2.38% on Friday, Oct. 9, released Microsoft-Decision-1. The model doesn't write open-ended text. It picks from a fixed set of options and gives each one a calibrated probability, so software can act on the result right away. The model is available now in Microsoft Foundry and will come to OpenRouter later. Microsoft says it beat both large language models and other decision models on speed and accuracy. The price is $0.042 per million input tokens, and output tokens are free.commandline.microsoft

Microsoft is offering the model as a cheap control layer for AI agents. These agents make many small choices between steps: which model to call, whether to retry, when to hand work to a person. Each of those choices adds delay and cost.testingcatalog+1

What Microsoft built

Chief Executive Satya Nadella announced the model on X. "It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality," he wrote. He added that Microsoft is "already testing it across Microsoft for everything from incident response and quality control to scientific discovery".seekingalpha+1

Microsoft built Decision-1 by further training Qwen3.5-9B, an open-weight model from Alibaba ↗ 5.36%, to score options in a single pass. The company plans to rebuild it on other base models, including its own MAI models and OpenAI's. It handles yes/no, multiple-choice and rating questions. It can also grade AI responses and agent actions against a rubric.commandline.microsoft

"Each decision adds delay, especially when one step depends on another," Achint Srivastava, a vice president in Microsoft's Office of the CTO, wrote in the announcement. He noted that adding 100 milliseconds to each of 20 decisions in a row adds two seconds to a workflow.commandline.microsoft

Company benchmarks

Microsoft says Decision-1 had the highest accuracy in its 36-benchmark comparison, which covered nearly 150,000 questions kept out of training. It was also the fastest model measured: 4.5 times quicker than the runner-up, Quyet-1.0-Large, and about 35 times quicker than OpenAI's GPT-6 Sol at median latency. Microsoft tested eight ways of rewording or reordering the same request, and the model changed its answer 1.3% of the time on average. Safety testing covered 5,250 requests across 11 benchmarks.commandline.microsoft

These results come from Microsoft's own testing and have not been checked independently.fourweekmba+1

Internal trials

Xbox Research used the model to sort more than 10,000 pieces of player feedback into themes. It reported quality comparable to GPT-6 Sol at more than 14 times the speed and one two-hundredth of the cost. Microsoft's Copilot team found it competitive with GPT-5.6 Luna and 100 times faster at judging response quality. In Microsoft Discovery, its scoring was 46 times more consistent than an approach based on a large language model.commandline.microsoft

Microsoft shares rose after the announcement, according to market data cited by Futu.news.futunn

"Now that agentic AI is a reality, we've seen that cost plays a major role in how people decide to use AI," Microsoft wrote.commandline.microsoft