Microsoft launches Decision-1 model to streamline AI decision-making
October 16, 2026 · 1 min read
Also published in Italiano, Português (Brasil)
Microsoft has launched a new AI model named Decision-1, designed to assess the likelihood of predefined answer sets. The model aims to alleviate workloads such as routing and prioritization, as well as direct result verification of other AI systems. This innovation allows applications to utilize numerical data for immediate execution, without relying on general large language models (LLMs) for redundant text processing.
The Decision-1 model has been developed through post-training on the foundational model Qwen3.5-9B and is available via Microsoft Foundry and OpenRouter. Microsoft plans to implement this process with its Microsoft AI (MAI) models and OpenAI models in the future.
Supported tasks for Decision-1 include determining the correctness of responses, with issues relayed to the relevant engineering teams. According to Microsoft, which tested the model against nearly 150,000 questions across 36 metrics, Decision-1 demonstrated the highest accuracy, outperforming the second-place model by a factor of 2.5 and exceeding GPT-6 Sol by 35 times.
In terms of pricing structure, Microsoft has set a service rate of 4.2 cents per 1 million input tokens, while no charges will apply for output tokens. However, these testing figures and costs are based solely on Microsoft's internal references, with no confirmation from independent testing agencies.