Microsoft Decision-1: 9B Qwen Model Launch Explained

shattered.io · Oct 26, 2026

Microsoft introduced Microsoft-Decision-1 on October 9, 2026, a specialized model that scores predefined answer choices rather than generating text. Built by post-training Alibaba’s 9-billion-parameter open-weight Qwen3.5-9B model, it returns calibrated probability scores for tasks including routing, classification, prioritization, verification and workflow control. It supports text inputs of up to 32,000 tokens but does not provide written explanations or process images, audio or video. Microsoft said future versions will be rebased on Microsoft AI and OpenAI models.

The model launched in public preview through Microsoft Foundry and became available on OpenRouter as microsoft/microsoft-decision-1. Its weights are not distributed, limiting use to hosted APIs. OpenRouter and secondary reports list pricing at $0.042 per million input tokens with no output-token charge. Microsoft said Decision-1 achieved 83.5% accuracy across 36 benchmarks containing roughly 150,000 held-out questions, and reported approximately 35 times lower P50 latency than GPT-6 Sol and 4.5 times faster performance than Quyet-1.0-Large; those results have not been independently reproduced. Microsoft also reported 98.7% perturbation stability and said it is testing the model internally for incident response, quality control and scientific discovery. Xbox Research reportedly used it to categorize more than 10,000 feedback items into predefined themes.

Read original ↗

What people are saying

The thread barely engaged with Microsoft Decision-1 as a model; it mostly treated the submitted write-up as untrustworthy AI-industry marketing.

What readers liked

Decision-focused models look useful when judged on concrete classification work such as phishing detection, rather than being asked to imitate…
Independent testers found some open alternatives compelling: pplx-decider performed well on MonitoringBench, while respan reportedly neared Jev…

Concerns raised

The launch coverage offers no comparison with Jev or other decision models, making Microsoft’s capability claims impossible to place against the…
Marketing-heavy, possibly machine-written prose turns a model announcement into unverifiable AI slop instead of evidence-based technical reporting.

View discussion ↗

Read full discussion analysis

Source: Hacker News · Based on 5 sampled comments · Analyzed Oct 26, 2026

Discussion overview

The thread barely engaged with Microsoft Decision-1 as a model; it mostly treated the submitted write-up as untrustworthy AI-industry marketing. Several commenters objected that a story about a decision model omitted the established Jev line and other competing approaches, leaving readers with no useful baseline for judging whether a 9B Qwen-derived model is novel, capable, or merely branded. The promotional tone and apparently synthetic prose made that omission feel worse: readers argued that unverifiable model claims dressed up as launch coverage are exactly how AI slop passes for technical reporting.

The substantive discussion shifted to what decision models can actually do and how they should be tested. One reader suggested a constrained vision-style classification task, while another cited phishing detection as a practical evaluation target. People who had run their own informal comparisons reported stronger impressions of alternatives: pplx-decider variants performed well in one benchmark, while an open respan model was said to approach Jev’s quality for less money. There was no firsthand evaluation of Decision-1 itself, so the prevailing conclusion was not that Microsoft’s model failed, but that the article supplied too little comparative evidence to make a case for it.

What readers liked

Decision-focused models look useful when judged on concrete classification work such as phishing detection, rather than being asked to imitate general-purpose chatbots.
Independent testers found some open alternatives compelling: pplx-decider performed well on MonitoringBench, while respan reportedly neared Jev quality at a lower cost.

Concerns raised

The launch coverage offers no comparison with Jev or other decision models, making Microsoft’s capability claims impossible to place against the tools practitioners already use.
Marketing-heavy, possibly machine-written prose turns a model announcement into unverifiable AI slop instead of evidence-based technical reporting.
No commenter supplied a firsthand Decision-1 benchmark, so its claimed value remains speculation while competing open models already have task-level reports.