Microsoft launched Decision-1, an AI model for routing, classification and verification. It returns probabilities for predefined answers instead of free-form text. It handles yes-or-no, multiple-choice, ratings and rubric assessments. It evaluates multiple decisions in parallel within one request. Microsoft claims it is 2.5 times faster than the next best model. It costs 4.2 cents per million input tokens with no output token charge. It is built by post-training Qwen3.5-9B and will be rebased on other models. Microsoft cites highest accuracy across 36 benchmarks with nearly 150,000 questions. Analysts note it can reduce cost and latency for agent workflows. Governance risks include lack of reasoning and sensitivity to question wording.

• 4.2 cents per million input tokens
• 2.5 times faster than runner-up
• 35 times faster than GPT-6 Sol
• 275,000 decisions per $1