Microsoft launches Decision-1, a structured-decision model built on Qwen3.5-9B
Microsoft released Microsoft-Decision-1 in Microsoft Foundry, a model built for structured decision scoring: given a fixed set of options it returns a calibrated probability for each, which software can consume directly for routing, classification, prioritization, verification, agent controls, data labeling and AI judging. It is post-trained from the open-weight Qwen3.5-9B, with OpenRouter availability planned. Microsoft says it ranked first in accuracy across 36 benchmarks covering nearly 150,000 questions, runs 35x faster than GPT-6 Sol at P50 latency, and is priced at $0.042 per million input tokens with output tokens free; Xbox Research reported labeling over 10,000 feedback items at comparable quality to GPT-6 Sol, more than 14x faster and about 200x cheaper. Microsoft says it plans to rebase the system on other models, including its MAI models and OpenAI technology.
Decision-scoring models are being pitched as a low-cost control layer for agentic workflows, and Microsoft's entry could speed up competition in that niche.