Microsoft said on Oct. 9 it has unveiled an AI model optimized for rapid decision-making, called Microsoft-Decision-1.

Decision models, unlike large language models (LLMs) that write text or reason through complex problems, produce only structured outputs that software can use immediately.

Microsoft-Decision-1 handles yes-or-no, multiple-choice and scoring requests. It also presents probability scores for each option. The company said it can be used for task classification, prioritization, verification and workflow control.

Microsoft said it developed the decision model by further training the open-source AI model Qwen3.5-9B. It plans to release versions based on its own MAI model and OpenAI models.

Microsoft also introduced internal applications. The Xbox research team used Microsoft-Decision-1 to classify more than 10,000 pieces of gamer feedback collected from surveys, Steam and X by topic, and it showed quality similar to GPT-6 Sol, it said.

It was more than 14 times faster and cost about one two-hundredth as much. The Copilot team used it to evaluate response quality, and it had quality similar to GPT5.6 Luna while being 100 times faster, it said.

Microsoft-Decision-1 can be used on Microsoft Foundry. It will also be available soon on OpenRouter. The price for input tokens is $0.042 per 1 million tokens, and output tokens are free.

Microsoft said, "In an era when agents act in the real world, decision models can help guide and control agents in complex environments."