fastino/GLiNER2.5-Decide — new model trending #30 on Hugging Face
Fastino's 340M GLiNER2.5-Decide classifier scores 60.2% exact match on a 17-domain benchmark.
Fastino published GLiNER2.5-Decide, a 340-million-parameter English classifier trending on Hugging Face that scores caller-supplied labels in one forward pass without prompts or generated tokens. On fastino/fast-decisions—17 domains with 300 held-out examples each—it averaged 60.2% exact-match accuracy, ahead of GLiNER2.5-Decide-1B at 59.6% and SemIf (Qwen3.5-4B) at 56.4%. It is meant for operational tasks such as intent, routing, sentiment, priority, moderation, and spam, and is loaded with AutoExtractor from the gliner2 package. Multilingual use is pointed to the 287M GLiNER2.5-multi-Decide model.
48