Efficient open model artifacts target production AI

NVIDIA's Nemotron 3.5 Lightning is an open-weights 30B MoE/Mamba hybrid with about 3B active parameters, long context, agent-focused deployment artifacts, and a vendor-reported 74% routing-cost reduction. Related lightweight releases include TokenAI Neo and Fastino GLiNER2.5-Decide, which offer open artifacts and low-latency decision inference but narrow or internally evaluated capabilities.