Fastino ships a 340M CPU decision model

Fastino released GLiNER2.5-Decide, a 340M-parameter open-weight model for schema-defined decisions that reports 167.3ms p50 latency on a 48-vCPU CPU. Its launch post reports 60.1% average accuracy across an internal 17-dataset suite, with Apache 2.0 weights that run locally on CPU or GPU. Builders routing tickets, tool calls, or handoffs should test this narrow encoder against their current frontier-model path before paying for generated tokens; the archive’s recent model-cost warning covers why workload output can erase token-price savings.