Microsoft-Decision-1 Claims 35x the Speed of GPT-6 Sol for $0.042 per Million Input Tokens, With Output Free

Microsoft joined the decision-model race on Friday. CEO Satya Nadella announced Microsoft-Decision-1 on X, describing a model purpose-built for structured decision tasks that, in the company's tests, "excels at structured decision tasks, outperforming large language models and other decision models on both latency and quality." Microsoft says the model has already been tested internally across incident response, quality control and scientific discovery scenarios.

The headline numbers, all from Microsoft's own benchmarks: Decision-1's P50 latency is 35 times faster than OpenAI's GPT-6 Sol, and 4.5 times faster than rival decision model Quyet-1.0-Large. The company says the model ranked first on accuracy across 36 benchmarks covering nearly 150,000 questions. None of these figures has been independently verified, and Microsoft's own materials note that the comparisons come from internal testing.

Unlike a chatbot, a decision model does not generate text. Given a set of predefined options, it picks the best one and attaches probability scores, letting downstream applications decide whether to proceed, retry, escalate, or hand the case to a human reviewer. Microsoft positions Decision-1 for routing, content classification, task prioritization, result verification and workflow control — the unglamorous choices that pile up inside agentic systems. The latency argument is arithmetic: if an agent makes 20 chained decisions and each adds 100 milliseconds, the workflow has quietly absorbed 2 seconds.

Pricing is the aggressive part: $0.042 per million input tokens, with output tokens free. That structure is aimed at high-frequency, repetitive decision workloads where per-call cost dominates. Microsoft says internal testing showed up to 200 times lower costs than comparable LLM-based approaches, another self-reported figure.

The company offered two internal case studies. Microsoft's Xbox research team used the model to analyze more than 10,000 customer comments and reviews, reporting results comparable to GPT-6 Sol at more than 14 times the speed and roughly 200 times lower cost. The Copilot team tested Decision-1 to evaluate AI-generated responses, saying it delivered competitive results at approximately 100 times the speed of another advanced AI model. Both examples carry the same caveat: they are Microsoft's own measurements.

Decision-1 is available through Microsoft Foundry, and the company says it plans to list the model on OpenRouter as well. The move lands in a category that has heated up fast: TypeSafe's Jev, which reportedly closed roughly $870 million at a $7.5 billion valuation this week led by Andreessen Horowitz; Cloudflare's open-source Clef; AWS's Strands Decider 2B; and OpenAI's limited-preview Decisions API all target the same layer. Microsoft is the largest platform company to ship a first-party entrant so far.

The broader signal is that decision models are moving from startup niche to platform feature, and the pricing column is where the fight will happen first. Independent benchmarks do not exist yet for any of the entrants, which means buyers are currently comparing vendor slide decks. For agentic workloads, where dozens of model calls sit between a user request and a finished task, the cost and latency of the decision layer may matter more than the flagship model's next benchmark score.