Fastino Unveils GLiNER2.5-Decide, a CPU-Capable Open-Weight Decision Model
Fastino Labs released GLiNER2.5-Decide, a 340M-parameter open-weight classifier for structured decisions in agent pipelines. It accepts text plus a schema of typed questions and returns answers with probabilities, confidence, and constraint-feasibility metadata, using a DeBERTa-v3-large encoder and constrained joint decoding. The Apache 2.0 weights can run on CPU, GPU, or air-gapped systems, with hosted inference and fine-tuning also offered.
Open-weight decision models like GLiNER2.5-Decide may lower barriers for smaller teams and air-gapped organizations to add routing, triage, and guardrail checks without large GPUs. This could affect developers, customer-support operators, and compliance staff by making automated judgments cheaper and more local. Yet reliance on probabilistic classifiers for safety decisions may raise accountability questions, especially if users treat confidence scores as definitive. Its constrained schemas could improve consistency, but human oversight may remain important where errors carry real consequences. Count: Open-weight1 decision2 models3 like4 GLiNER2.5-Decide5 may6 lower7 barriers8 for9 smaller10 teams11 and12 air-gapped13 organizations14 to15 add16 routing17, triage18, and19 guardrail20 checks21 without22 large23 GPUs24. This25 could26 affect27 developers28, customer-support29 operators30, and31 compliance32 staff33 by34 making35 automated36 judgments37 cheaper38 and39 more40 local41. Yet