GLiNER2.5-Decide launches with structured outputs, a fine-tuning API and hosted inference
Its developers say the model goes beyond picking labels: it can extract text spans, identify relationships and return structured records, with rules to keep related decisions consistent.
TLDR
The developers announced GLiNER2.5-Decide alongside a new API for fine-tuning GLiNER models directly inside a coding agent and a hosted inference option. They say the model can apply exclusions, limits and other rules across related decisions. For short, single-document requests, they report 167 ms CPU latency on a 48-vCPU Intel Xeon Platinum 8581C and 38–47 ms GPU latency on T4, L4, V100 and A100 hardware. They also shared model weights on Hugging Face.
