On October 9, 2026, Microsoft launched Decision-1, a specialized model for routing, classification, and rubric-based evaluation that returns calibrated probabilities for yes/no, multiple-choice, and rating-scale tasks. Microsoft claims the model delivers the highest accuracy across 36 benchmarks covering nearly 150,000 questions, runs 4.5 times faster than Quyet-1.0-Large and 35 times faster than GPT-6 Sol, and produces stable decisions under input perturbation 98.7% of the time. It is priced at $0.042 per million input tokens with free output, is based on Qwen3.5-9B, and is available through Microsoft Foundry and OpenRouter.
