Microsoft releases AI decision model, claiming 35-times faster responses than GPT-6 Sol
Decision-1 is available in Foundry for routing, classification and workflow checks. Its speed and accuracy pitch rests on Microsoft’s own structured-task benchmarks.
Microsoft’s October 9 launch introduces Decision-1 for producing structured outputs that applications and agents can act on, rather than prose for users. In the company’s evaluation of 36 benchmarks and nearly 150,000 questions, the model ranked highest in accuracy and was reported 35× faster than GPT-6 Sol and 4.5× faster than Quyet-1.0-Large. The findings apply to the structured tasks tested, not open-ended writing or complex reasoning. Decision-1 is available through Microsoft Foundry; OpenRouter access is planned, not yet live.
01
Microsoft lists routing, classification, prioritization, verification and workflow control as intended uses.
02
Microsoft says the benchmarks were kept blind from training, but the results remain company-reported comparisons.
03
Achint Srivastava, Microsoft’s vice president of software engineering in the Office of the CTO, authored the launch announcement.
Microsoft released Microsoft-Decision-1 on October 9, 2026, giving developers a dedicated model for routing, classification and workflow checks in Microsoft Foundry. The launch targets decisions inside applications and AI agents. Microsoft says the model achieved the highest accuracy and lowest response delay among the systems in its own structured-task comparison.
A decision, not a written response
The distinction is the output. Microsoft describes language models as systems designed to generate text or reason through complex problems. Decision models instead deliver structured outputs that software can immediately act on. Decision-1 is built for that narrower role: the result serves the application’s next decision, rather than supplying prose for a person to read.
Microsoft names five intended uses: routing, classification, prioritization, verification and workflow control. That spans directing work, assigning categories, setting priorities and checking results. The company is positioning Decision-1 as a component inside existing applications, agents and workflows, rather than a standalone assistant.
The accuracy claim has a defined scope
Microsoft’s comparison covered 36 benchmarks and nearly 150,000 questions. The company says the benchmarks were kept blind from training, and that Decision-1 achieved the highest accuracy among the systems compared. The evaluation therefore pairs a speed claim with a separate claim about answer quality.
These are company-reported results for structured decision tasks. They should not be read as evidence that Decision-1 outperforms the compared models at open-ended writing or complex reasoning. The relevant comparison is how well—and how quickly—the systems handle the bounded tasks Microsoft evaluated.
Foundry now, OpenRouter later
Microsoft Foundry is the access route at launch. Microsoft also says the model is coming soon through OpenRouter, making that a planned distribution channel rather than current availability. Achint Srivastava, Microsoft’s vice president of software engineering in the Office of the CTO, authored the launch announcement.
Reader comments
Newest comments first. Replies stay oldest first.