Microsoft unveils Microsoft-Decision-1, a decision model built on Alibaba's Qwen

Microsoft has launched Microsoft-Decision-1, a small model designed to choose among a fixed set of answers rather than write prose. According to the Foundry model card, it was built on Alibaba's open-weight Qwen3.5-9B and then post-trained. Achint Srivastava, a vice president of software engineering in Microsoft's Office of the CTO, announced the release in a blog post on 9 October.

Srivastava wrote that Microsoft will soon rebase the model on other systems, including OpenAI's models and Microsoft's own MAI models, though the post does not explain the reasoning. He framed the category as fast-growing, writing that decision models are quickly emerging as an important new category in AI.

Given a fixed set of options, the model returns a calibrated probability score for each one, handling yes-or-no questions, multiple choice and ratings. Microsoft says it can also grade AI responses and agent actions against a rubric, and lists routing, classification, prioritisation, verification and workflow control among its uses. The model accepts text only and has a 32,000-token context window; the model card warns it should not be the sole basis for consequential decisions about people, such as credit or employment.

Microsoft said the model posted the highest accuracy in its own 36-benchmark comparison, which covered nearly 150,000 questions kept blind from training. In those tests it ran 2.5 times faster than the runner-up, H2O-Lightning-4B v1.1, and 35 times faster than OpenAI's GPT-6 Sol. When the same request was reworded or reordered, the model changed its decision 1.3% of the time on average, and Microsoft also tested it on 5,250 requests spanning 11 safety benchmarks covering harmful content, jailbreak attempts and prompt injection.

Internal teams have already put the model to work. Xbox Research used it to sort more than 10,000 pieces of player feedback and reviews into themes, and Microsoft said it was competitive with GPT-6 Sol on quality while running more than 14 times faster and 200 times cheaper. The Copilot team found it competitive with GPT5.6 Luna and 100 times faster, according to the company.

Input costs $0.042 per million tokens and output tokens are free. The model is available in Microsoft Foundry and through OpenRouter, the model router Stripe acquired in August. Microsoft later updated its post to add benchmarks for Jev from TypeSafe AI, which launched three weeks earlier and became the fastest-adopted model on Vercel this year, The Register reported; Microsoft said its model was 2.8 times faster than Jev in its latency test. OpenAI's Decisions API is in public beta, and Cloudflare, Strands, Liquid AI, Perplexity and Snowflake have also announced decision models, with more than 100 now competing for attention, according to The Register.