New to this? Read it in simple words
- Microsoft released a small model that picks answers from a fixed list.
- It is built on Alibaba’s open Qwen3.5-9B model.
- It costs $0.042 per million input tokens, the same as Jev.
- Microsoft says it beat rivals on 36 tests; nobody else has checked yet.
- Decision model
- An AI model that scores a fixed list of options instead of writing text.
- Rebase
- To rebuild a model on top of a different base model.
- Token
- A small piece of text, often part of a word, that AI models read and are paid by.
A decision model does not write text. It gets a question and a fixed list of answers, and returns how likely each answer is. Since TypeSafe launched Jev in September, several large companies have built their own.
What Microsoft released
Microsoft announced Microsoft-Decision-1 on its Command Line blog on October 9. It handles yes-or-no questions, multiple choice and ratings, and it can grade AI answers and agent actions against a rubric.
Microsoft says it trained the model further from Alibaba’s open Qwen3.5-9B. The model reads only text, up to 32,768 tokens, and answers in a fixed data format.
It costs $0.042 per million input tokens, and the output is free. The New Stack points out that this is exactly what TypeSafe charges for Jev. Microsoft has not announced open weights.
What the model is built on, what it reads, what it costs, and what Microsoft claims, from Microsoft’s announcement and Foundry guide.
Microsoft’s own test results
Microsoft says the model had the highest accuracy across 36 tests with nearly 150,000 questions. The Register gives its score as 83.5%.
Microsoft also says the model is 2.5 times faster than the runner-up, H2O-Lightning-4B, and about 35 times faster than OpenAI’s GPT-6 Sol. All of these are Microsoft’s figures.
Microsoft’s own guide is more careful. It tells customers to check the scores on their own labelled examples, and to use them for ranking options rather than as exact ratings.
Next: other models underneath
Microsoft “will soon rebase it on other models, including Microsoft AI (MAI) and OpenAI,” the announcement says. The Register notes that Microsoft did not say why.
The New Stack and The Decoder count Microsoft as one of several companies with a decision model since mid-September. Cloudflare and AWS released theirs in the past weeks.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · October 9, 2026Introducing Microsoft-Decision-1, our model for fast decision-making Microsoft
- 2Primary source · October 9, 2026Deploy and use Microsoft-Decision-1 in Microsoft Foundry Microsoft Learn
- 3
- 4Research · October 10, 2026Microsoft leans on open weight model from Chinese AI lab to challenge Jev The Register
- 5Research · October 10, 2026Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen The New Stack
- 6Research · October 10, 2026Microsoft’s Decision-1 model enters the fast-growing AI decision model race The Decoder
Every test result here is Microsoft’s own, and no outside group had tested the model by October 11. Microsoft updated its announcement after launch, and we use the current version.
Microsoft launched a cheap, fast decision model on Alibaba’s Qwen, but its winning test results are still only Microsoft’s own claims. How we rate the mood