Microsoft launched Decision-1, a narrow AI routing model that scores fixed options instead of generating free-form answers. The Microsoft announcement says it is available through Microsoft Foundry and OpenRouter.
Post-trained on Qwen3.5-9B, the model returns calibrated probabilities for choices such as yes/no decisions, ratings and rubric-based grades. Input costs $0.042 per million tokens, while output tokens are free. Microsoft says Decision-1 led its 36-benchmark comparison across nearly 150,000 blind questions, with P50 latency about 35x faster than GPT-6 Sol and 4.5x faster than Quyet-1.0-Large. Those performance results are vendor-reported.
For builders, the practical play is replacing some LLM judge calls with cheaper probability scoring, then using confidence thresholds to route requests, trigger review or escalate uncertain cases.