The model card says callers can supply label sets at runtime for intent, routing, sentiment, priority, policy, and multi-label tagging, with several decision heads evaluated in one forward pass. It is positioned for local deployment under Apache-2.0 rather than as a general-purpose answering model.
On Fastino’s English fast-decisions benchmark—17 domains with 300 held-out examples per domain—the model card reports 60.2% average exact-match accuracy for GLiNER2.5-Decide. The card explicitly limits the model’s role: it does not reason, explain, or answer open questions. A separate multilingual variant is recommended for multilingual input.