Microsoft released Microsoft-Decision-1, a decision-scoring model. It outperforms LLMs and other decision models on latency and quality. The model is available in Microsoft Foundry and via OpenRouter. It supports yes/no, multiple-choice, rating, and rubric-based grading tasks. Microsoft-Decision-1 achieved the highest accuracy in a 36-benchmark test. It was 2.5 times faster than H2O-Lightning-4B v1.1 and 35 times faster than GPT-6 Sol. Input tokens cost $0.042 per million, with output tokens free. It maintains a 1.3% decision flip rate under input perturbations. The model is designed for routing, classification, prioritization, and workflow control. Documentation and pricing details are published.
What changed
Decision tasks relied on general-purpose LLMs or slower decision models.
What it unlocks
Running fast, low-cost decision scoring for routing, classification, and workflow control in applications.
- 2.5x faster than H2O-Lightning-4B v1.1
- 35x faster than GPT-6 Sol
- $0.042 per million input tokens
- 1.3% decision flip rate on perturbations
What you need to act on it
- access to Microsoft Foundry or OpenRouter
Sources