Microsoft's new decision model is Alibaba's Qwen underneath
On 9th Oct'26, Microsoft released Microsoft-Decision-1, a small model that chooses between answers you supply, such as which team gets a ticket or whether an agent's action passes a rubric, and returns a probability for each. It is in Microsoft Foundry and, Microsoft says, on OpenRouter, at $0.042 per million input tokens with output free. Microsoft's announcement says it built the model by further training Alibaba's open-weight Qwen3.5-9B, and that it plans to rebase it on Microsoft AI and OpenAI models.
What it actually means
We covered Cloudflare's version of this on 6th Oct'26, and the new fact is not the price but the parts list. A product Microsoft sells through its own cloud is a Chinese open-weight model underneath, and Microsoft said so in its own announcement. That answers part of the question teams keep asking about open weights from China: once the weights are downloaded and retrained, the model runs where you put it, and what you are buying is the vendor who stands behind it.
The catch is that procurement teams in Gulf banks and European ministries often ask where a model came from as well as where it runs, and this one will give a different answer once Microsoft rebases it. If you route decisions through it, log which version made each call and keep your own set of labelled cases to rerun when the base model changes, because the accuracy figures so far are Microsoft's own and nobody outside has tested them.
The model a vendor sells you may not be the vendor's model.
$0.042per million input tokens for Microsoft-Decision-1, output free
Source: Microsoft Command Line · Monday 12 October 2026