Microsoft-Decision-1 decision model
Confirmed
The takeaway
On Oct 9, 2026 Microsoft introduced Microsoft-Decision-1, a decision-scoring model for routing, classification, prioritization, verification and workflow control, post-trained from Qwen3.5-9B.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 10 October 2026
Your AI and this story
- GPT-6 Astra162 days after its cutoff
- Claude Opus 5.5101 days after its cutoff
- Gemini 3.8 Flash192 days after its cutoff
- Grok 4.7131 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 101 days before it.
Key facts
- Announced Oct 9, 2026 on Microsoft’s Command Line blog (‘Introducing Microsoft-Decision-1, our model for fast decision-making’) by Achint Srivastava, VP of Software Engineering, Office of the CTO
- Base: post-trained Qwen3.5-9B; Microsoft plans to rebase it on other models, ‘including Microsoft AI and OpenAI’
- Microsoft’s claims: highest accuracy in a 36-benchmark comparison spanning ~150,000 questions; p50 latency ‘2.5 times quicker than H2O-Lightning-4B v1.1, the runner-up, and 35 times quicker than GPT-6 Sol’; changes its decision on 1.3% of perturbations on average; refused harmful behaviour on 5,250 requests across 11 benchmarks. No per-benchmark accuracy numbers were published in the post
- Internal users (Microsoft’s figures): Xbox Research ‘14 times faster and 200 times less expensive’ than GPT-6 Sol; Copilot team ‘100 times faster’ than GPT-5.6 Luna; Microsoft Discovery ‘46 times more consistent’ scoring
- Price: $0.042 per 1M input tokens, output free; available in Microsoft Foundry (catalog id Microsoft-Decision-1) and, per Microsoft, OpenRouter (the OpenRouter page returned 404 and the model was absent from OpenRouter’s models API on Oct 10)
- No weights release is mentioned in the post (no Hugging Face repo found on Oct 10)
- Hacker News (185 points): a commenter reported it as slower than Jev at p50 (figures of 283 ms and 369 ms were quoted); unverified community measurement
What happened
Microsoft published Microsoft-Decision-1 on Oct 9, 2026, the same day TypeSafe announced its $870M Series A. Like TypeSafe’s Jev, Cloudflare’s Clef, Perplexity’s pplx-decider and OpenAI’s Decisions API, it is a “decision model”: it scores a fixed set of options for typed questions (routing, classification, prioritization, verification) and returns a choice instead of generating free text. It is a post-trained Qwen3.5-9B, and Microsoft says it will rebase it on Microsoft AI (MAI) and OpenAI models later.
All accuracy, latency and robustness figures are Microsoft’s own; the post gives ratios and totals but no per-benchmark scores.
Why it matters
Another large company has shipped a decision model priced like Jev within four weeks of Jev’s launch, which confirms decision models as a product category. Microsoft basing it on a Chinese open-weight model (Qwen) rather than on its own MAI or OpenAI models is notable.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 2 primary, 1 reaction
Primary
- Microsoft Command Line: Introducing Microsoft-Decision-1, our model for fast decision-makingcommandline.microsoft.com, official
- Microsoft Foundry model catalog: Microsoft-Decision-1ai.azure.com, docs
Reactions
- Hacker News discussion (185 points)news.ycombinator.com, discussion
Changes
- Filed