GLiNER2.5-Decide: Classification, Routing, and Sentiment Analysis in One with a 340M Model

GLiNER2.5-Decide is a lightweight 340M parameter model that instantly classifies text without prompt templates or token generation. Label sets can be dynamically specified at call time, allowing various tasks such as intent classification, sentiment analysis, and document type determination to be handled in a single inference pass.

It is specialized for repetitive decision-making tasks in operational environments, such as customer support ticket routing, banking request classification, and medical appointment sorting. Multiple heads, including intent, urgency, and responsible department, can be scored simultaneously in a single call, simplifying pipelines and reducing latency.

It achieves an average accuracy of 60.2% across 17 domain benchmarks, demonstrating efficient performance compared to competing models with larger parameter counts. Optimized for English input, it is available under the Apache 2.0 license for local deployment, making it highly suitable for environments where data security is critical.