Microsoft introduced Decision-1 on October 9, 2026, a model that refuses to chat or generate text. Instead, it scores a fixed list of answer options in a single forward pass, returning a probability for each choice. Built on Alibaba's Qwen3.5-9B, it targets internal uses like incident response and scientific discovery, bypassing the need for open-ended generation.
The model is available immediately through Microsoft Foundry and Open Router. Pricing is set at $0.042 per million input tokens, with free output availability. This narrow, task-specific design distinguishes it from the industry trend of building larger, chattier assistants, offering a cheaper, faster alternative for routing, classification, prioritization, and verification tasks.
Early internal trials suggest strong potential for engineering teams building agent pipelines and automated verification systems. While the launch leverages a Chinese AI lab's open-weight model, Microsoft plans to eventually rebase the model away from Qwen dependencies. For builders focused on production efficiency over general capabilities, Decision-1 represents a strategic shift toward specialized, calibrated judgment rather than raw generative power.
Source: Hacker News · Summarized by HeadlinesBriefing