Microsoft launches Decision-1, a fast decision-scoring model built on Alibaba's Qwen3.5-9B

The 9-billion-parameter model returns probabilities for fixed answer options, with Microsoft planning to rebase it on its own models and OpenAI

By LineZotpaper AI staff

Published26 October 2026

Updated26 October 2026

Read Time2 min

Follow this storyHear when it moves

Illustration: Microsoft launches Decision-1, a fast decision-scoring model built on Alibaba's Qwen3.5-9B

On October 9, 2026, Microsoft introduced Microsoft-Decision-1, a 9-billion-parameter model that scores a fixed list of answer options in a single forward pass and returns a probability for each one, rather than generating text. Built by post-training Alibaba's open-weight Qwen3.5-9B, the model targets routing, classification, prioritisation and verification work, and is available through Microsoft Foundry and OpenRouter.

Microsoft says the model is designed for high-volume, low-latency decisions such as routing a support ticket, making a yes/no moderation call, or checking an AI agent's output before it proceeds. It was made available through Microsoft Foundry immediately and through OpenRouter shortly after, with internal uses ranging from incident response to scientific discovery.

The Register framed that choice as Microsoft leaning on a Chinese AI lab's open-weight release to get the product out the door quickly, while noting Microsoft's stated intent to migrate the foundation over time.

Microsoft supports the launch with vendor-reported numbers. It says Decision-1 achieved the highest accuracy across 36 benchmarks and about 150,000 held-out questions, at 83.5 per cent accuracy. It also reports that the model runs 4.5 times faster than Quyet-1.0-Large, 2.5 times faster than H2O-Lightning-4B v1.1, and shows approximately 35 times lower P50 latency than GPT-6 Sol on its own benchmark. Taiwan's Liberty Times reported response times of about 85 milliseconds, without specifying hardware or test conditions. Microsoft also says the model's decisions were perturbation-stable 98.7 per cent of the time, meaning small rewording of input did not flip the selected option.

None of the coverage reviewed for this piece includes an independent third-party reproduction of the 36-benchmark comparison, the speed multipliers or the latency figure, so those numbers are best read as vendor claims. Pricing circulating from OpenRouter's model listing, picked up by AI Weekly and the Italian outlet Pasquale Pillitteri, is $0.042 per million input tokens with output free, a structure that fits a model whose output is only a short list of probability scores.

RSS

The Daily Edition

The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.

Analysis

Why This Matters

Microsoft is betting that many production systems need fast, cheap, low-latency decisions rather than another general-purpose chatbot, and is pricing the model accordingly.
The choice of Alibaba's open-weight Qwen3.5-9B as a base shows that even a company with its own AI ambitions will use a rival's open weights to get to market quickly.
For engineering teams, a model with a fixed output format could simplify agent pipelines, content moderation and automated verification.

Background

Microsoft has invested heavily in its own model development through Microsoft Foundry and in OpenAI through its partnership, which makes the Qwen base notable even as a starting point. The company says the underlying engine will change over time, moving to Microsoft AI and OpenAI models. Building a task-specific tool on a compact open-weight model is not unusual in the industry, but naming the product Decision-1 signals an attempt to make decision scoring a distinct product category rather than a general assistant. Qwen is an open-weight model family from Chinese technology company Alibaba.

Key Perspectives

Microsoft: says Decision-1 is built for routing, classification, prioritisation, verification and workflow control, and describes Qwen3.5-9B as a starting point. In its launch post: “To build Microsoft-Decision-1, we post trained Qwen3.5-9B for fast, single-pass decision scoring and will soon rebase it on other models, including Microsoft AI (MAI) and OpenAI.” The Register: framed the launch as Microsoft leaning on a Chinese AI lab's open-weight release to get the product out quickly, while noting Microsoft's stated intent to migrate the foundation over time.

What to Watch

Whether Microsoft follows through on its stated plan to rebase Decision-1 on Microsoft AI and OpenAI models, and on what timeline.
Whether anyone outside Microsoft independently reproduces the benchmark, speed and latency claims, particularly the 98.7 per cent perturbation-stability figure.
Whether the $0.042-per-million-input-token pricing attracts high-volume routing and moderation workloads away from general-purpose models.

Sources

Microsoft Decision-1: 9B Qwen Model Launch Explained · Hacker News

Checked

12 claims in this story checked against 1 source: 1 supported.

Show each claim

SupportedMicrosoft launched Decision-1.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesDecision-1 is a fast model.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Partly supportedDecision-1 was built on Alibaba’s Qwen3.5-9B.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesMicrosoft introduced Microsoft-Decision-1 on October 9, 2026.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesThe model scores a fixed list of answer options in one forward pass and returns a probability for each option instead of generating text.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesMicrosoft plans to rebase the model on its own models and OpenAI.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesThe model is available through Microsoft Foundry and OpenRouter.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesMicrosoft reports 83.5% accuracy, the highest across 36 benchmarks and about 150,000 held-out questions.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesDecision-1 runs 4.5 times faster than Quyet-1.0-Large and 2.5 times faster than H2O-Lightning-4B v1.1.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesThe model has approximately 35 times lower P50 latency than GPT-6 Sol on Microsoft’s benchmark.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesThe model’s decisions were perturbation-stable 98.7% of the time.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”
Not in the sourcesOpenRouter’s listing prices input at $0.042 per million tokens and makes output free.Hacker News: Front Page: “Microsoft Decision-1: 9B Qwen Model Launch Explained.”

Correction

26 October 2026An earlier version of this story presented wording as a direct quotation from Microsoft's launch post. The cited source does not substantiate that quotation, and it has been removed.

Discussion elsewhere

Hacker News71 points · 31 comments →

Reads like an ad for Microsoft and its model because it's an article about a decision model that doesn't mention Jev or other decision models at all.
mstkllah on Hacker News

Unlike the Internet, Microsoft did recognise the pivotal moment. But they made the wrong play… that and they’ve annoyed pretty much everybody with their products, I don’t think I’ve ever seen such a successful company with such little good will from their community.
yakbarber on Hacker News