Microsoft launches Microsoft-Decision-1, a decision model built on Qwen3.5-9B that picks from predefined options at $0.042 per million input tokens

Why it matters

【Key points】

At 02:37 Beijing time on Oct 10, Microsoft CEO Satya Nadella introduced Microsoft-Decision-1, a model for fast decision-making. The Microsoft Foundry blog put it in public preview in the Foundry Models catalog.

It does not write open-ended text. It picks among options the developer defines and returns answers plus confidence signals, for classification, model routing, data labeling, AI judging, incident routing and agent workflow control.

Microsoft says it is built on Alibaba's open-weight Qwen3.5-9B; future versions will be rebased on OpenAI and Microsoft MAI models. Price: $0.042 per million input tokens (US and EU Datazones); output is not billed.

【Verifiable】

Nadella's post: https://x.com/satyanadella/status/2108627923888754862

Foundry blog: https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562742

The blog says Xbox Research sorted 10,000+ feedback items with quality competitive with GPT-5 at 80–100x the speed. X claims such as "35x faster than GPT-6 Sol" are not in the blog text. About 75 minutes in, Nadella's post had ~175K views, and 300+ X posts mentioned the model.

【Our read】

1. Decision models are becoming their own category. Most calls in an agent system are small judgments, not long-form writing, and a 9B specialist handles them far cheaper and faster than a frontier model.

2. A new model on Microsoft's cloud is openly built on Chinese open weights: base-model choice follows performance and licensing, not origin. With a move to OpenAI and MAI bases planned, Qwen looks like a bridge.

3. Accurate multiple choice is not safe automation. Calibrated confidence, and not forcing a pick when no option fits, decide whether it can run a workflow; Microsoft itself says to validate on your own data.