Microsoft-Decision-1 AI Model Is Here With 35x Faster Performance Than GPT-6 Sol

Microsoft has launched the Microsoft-Decision-1, a new AI model designed to make structured decisions faster and at a lower cost than conventional large language models. The latest model targets tasks such as classification, prioritization, AI response evaluation, and controlling agent workflows. Microsoft claims it delivers the highest accuracy across its 36-benchmark evaluation while outperforming competing models in speed.

Microsoft-Decision-1 targets AI tasks that need quick decisions

Unlike traditional large language models that generate text and handle complex reasoning, Microsoft-Decision-1 focuses on selecting the best option from a predefined set of choices. It can assign probability scores to possible answers, helping applications decide whether to proceed, retry, escalate a task, or request human review.

Microsoft says the model achieved the highest accuracy across 36 benchmarks covering nearly 150,000 questions. It also reportedly ran 4.5 times faster than Quyet-1.0-Large and 35 times faster than GPT-6 Sol in the company’s tests. These figures come from Microsoft’s own benchmarking, so independent testing will be important to verify the claims.

Microsoft claims major gains in speed and cost

Microsoft-Decision-1 is built on Qwen3.5-9B through additional post-training focused on single-pass decision scoring. Microsoft also plans to rebase the model on other systems, including its MAI models and OpenAI models.

The company adds that its internal teams have already tested the model across several workloads. Xbox Research used it to categorize more than 10,000 pieces of gaming feedback, reporting comparable quality to GPT-6 Sol at over 14 times the speed and 200 times lower cost. Microsoft’s Copilot team also found it competitive with GPT-5.6 Luna for evaluating AI responses.

Developers can try Microsoft’s new AI model

Microsoft-Decision-1 is available through Microsoft Foundry, with OpenRouter support planned for later. Pricing starts at $0.042 per million input tokens, while output tokens are free, potentially making it attractive for applications that perform frequent, repetitive decisions.