Nvidia launches open-weight Nemotron 3.5 Lightning, eyes trillion-parameter rival to Chinese models

Key Points

• Nvidia released Nemotron 3.5 Lightning, a 30B‑parameter MoE model with a 1M token context window.
• The model ships under the OpenMDW‑1.1 license, allowing free commercial use and modification.
• Lab tests show token‑generation speed up to four times faster than comparable models.
• In real‑world benchmarks, Nemotron 3.5 completed 10,000 tasks 30% faster than Alibaba’s Qwen 3.6 35B.
• NeMo Switchyard routes tasks to the most suitable model, cutting inference costs by up to 74%.
• Nvidia plans a trillion‑parameter Nemotron 4, aiming for a late‑autumn release.
• Even at a trillion parameters, Nemotron 4 would still be smaller than China’s Kimi K3.
• The release reinforces Nvidia’s strategy of using open models to drive GPU demand.