Microsoft has launched Microsoft-Decision-1, a decision-making model post-trained from Qwen3.5-9B, on OpenRouter. It reports the highest accuracy across 36 blind benchmarks (~150K questions), running 4.5x faster than the runner-up and 35x faster than GPT-6 Sol, with a decision flip rate of just 1.3% on perturbed inputs. Input is priced at $0.042/M tokens, output is free, and context length is 32K. Nadella says it is already used internally for incident response, quality control, and scientific discovery.