Nvidia releases fast Nemotron 3.5 Lightning open-weights model
Nvidia has released Nemotron 3.5 Lightning, the first model in its Nemotron 3.5 line. It uses a hybrid Mamba-Transformer architecture with 31.6 billion total parameters, while only 3.6 billion are active at each step. The reasoning model processes text only and supports a context window of one million tokens.
Artificial Analysis gives Lightning an Intelligence Index score of 24, nine points above Nemotron 3 Nano 30B A3B and equal to OpenAI’s gpt-oss-120b. Smaller models including Qwen3.6 35B A3B and Meta’s Muse Glimmer score higher, while Nvidia’s Nemotron 3 Super reaches 26.
In pre-release tests with final NVFP4 weights, Lightning produced almost 670 tokens per second, ahead of the compared models and nearly twice as fast as Gemini 3.5 Flash-Lite. It scored 824 Elo on GDPval-AA v2, surpassing gpt-oss-120b and Nemotron 3 Super, and achieved 24.3 percent on Terminal-Bench v2.1. Nvidia offers BF16 and NVFP4 weights under the OpenMDW-1.1 license, with inference available through several cloud providers.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.