Microsoft takes Qwen 9B and modifies it into a decision-making model, self-tested to be 35 times faster than GPT-6 Sol.

Beating AI News Flash: Microsoft releases Microsoft-Decision-1, specifically designed to help AI agents make rapid judgments. It is post-trained based on Qwen3.5-9B, scoring only preset options and returning probabilities for each. For example, deciding which model a user request should be routed to, or determining whether an agent should continue, retry, or escalate to a human.

Microsoft claims that Decision-1 ranks first in accuracy across 36 evaluations and nearly 150,000 questions. It is 4.5 times faster than the second-place Quyet-1.0-Large and approximately 35 times faster than GPT-6 Sol.

In internal trials, the Xbox research team used it to categorize over 10,000 pieces of player feedback. Microsoft says the judgment quality is close to GPT-6 Sol, over 14 times faster, and costs only about 1/200 of the latter. The Copilot team also uses it to score AI responses.

Decision-1 is now available on Microsoft Foundry and OpenRouter, and Vercel AI Gateway has also integrated it. It is priced at $0.042 per million input tokens, with output free of charge. Microsoft has not yet announced that it will open-source Decision-1's model weights.