Nemotron 3.5 Lightning Hits Merge Gateway

Nemotron 3.5 Lightning makes open-weight inference more practical by optimizing for sustained speed rather than sheer parameter count.

• –Sparse activation should reduce serving costs and latency for high-volume agent tasks
• –A 1M-token context window enables long-running workflows and large RAG workloads
• –Merge Gateway availability removes deployment friction for developers testing the model
• –NVIDIA’s claimed 4x output-rate advantage and 86% PinchBench score are promising, but independent evaluations will matter
• –The model’s value will depend on tool calling, reliability, and quality under concurrency—not just benchmark throughput