Microsoft releases Microsoft-Decision-1 decision-scoring model

On October 9, 2026, Microsoft released Microsoft-Decision-1, a decision-scoring AI model post-trained from Alibaba's Qwen3.5-9B. Designed to score closed answer options rather than generate prose, the model achieved 83.5% accuracy across 36 benchmarks with an 85 ms median latency. Available on Microsoft Foundry and OpenRouter at $0.042 per million input tokens, it restricts ungrounded outputs but faces disputed latency comparisons from rival developer H2O.ai.

Pricing and API deployment

• ▪On Microsoft Foundry and OpenRouter, Microsoft-Decision-1 costs $0.042 per million input tokens with free output tokens, versus OpenAI's Luna decisions endpoint at $0.10 per million with free output tokens
• ▪When deployed on OpenRouter, Microsoft-Decision-1 operates via the Decisions API rather than standard chat endpoints, making chat completions software development kits incompatible

Model limitations and requirements

• ▪Microsoft specified that applications utilizing Microsoft-Decision-1 must establish defined decision options, score thresholds, escalation paths, and mechanisms for human oversight
• ▪Microsoft-Decision-1 outputs do not include explanations or rationales, and the model is prohibited from making sole automated decisions on credit, employment, housing, healthcare, or legal rights

Debatable claims

• ▪Decision scoring without output rationales poses unacceptable risks for automated AI systems
• ▪Microsoft's closed release of Microsoft-Decision-1 undermines the open-source AI ecosystem