Fastino unveils GLiNER2.5-Decide: a high-performance decision model for operational tasks

Fastino Labs has introduced GLiNER2.5-Decide, a 340M-parameter open-weight decision model designed for operational decision-making tasks. The model processes text and a schema of typed questions to generate structured answers, complete with probability distributions, confidence scores, and constraint-feasibility metadata. It is optimized for deployment on CPUs, GPUs, or in air-gapped environments, with hosted inference and fine-tuning available via the GLiNER API. GLi,NER2.5-Decide is a non-generative classifier based on the DeBERTa-v3-large encoder, fine-tuned from gliner2-large-v1. It excels in tasks such as routing, triage, tool selection, and guardrails, with a focus on operational decisions rather than reasoning or open-ended questions. The model achieved strong performance on Fast Decisions, an internal benchmark suite, outperforming other models in 9 out of 17 datasets, including customer operations and domain routing. Fastino also released GLiNER2.5-Decide-1B and GLiNER2.5-multi-Decide for multilingual use cases, with the 340M model leading in intent routing and scoring 75.3% on support intent. The model's efficiency is highlighted by its low-latency performance, with p50 latency at 52.6 ms on an A100 GPU for 1,024 tokens.