Nemotron 3.5 Lightning
A Mixture-of-Experts (MoE) large language model released by NVIDIA, with approximately 30 billion total parameters of which only about 3 billion are activated per token. This architecture maintains substantial knowledge capacity while achieving high inference efficiency.
Core Facts
Nemotron 3.5 Lightning is a 30B-scale Mixture of Experts (MoE) model that activates only approximately 3B parameters per inference.
Cline CLI v3.0.74's Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1, and removes Mimo V2.6 Flash and Muse Spark 1.3 Contributor.
Timeline (last 90 days)
Samaya AI released Sterling, a post-trained model for financial research scenarios, built on Nemotron 3.5 Lightning with approximately 3B activated parameters.
The two models share the same training philosophy: large-scale distillation, long-context training, and native speculative decoding.
Super is a 120B-parameter, 12B-activated-parameter multimodal model; Lightning is a 30B-parameter, 3B-activated-parameter text-only model.
Both models use a mixture-of-experts (MoE) and hybrid Mamba-Transformer architecture, targeting reasoning, long-context understanding and agentic workloads.
Cline CLI v3.0.74's Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1, and removes Mimo V2.6 Flash and Muse Spark 1.3 Contributor.
The video mentioned that the technical details of Nemotron 3.5 Lightning will also be included in the technical report, showing that this generation of the product line has more than one model.
Nemotron 3.5 Lightning is positioned for long-running agents, sub-agent collaboration, chatbots, and RAG systems, rather than one-off question answering.
Nemotron 3.5 Lightning uses a hybrid structure that interleaves Mamba-2 with MoE layers, paired with a small number of attention layers.
Nemotron 3.5 Lightning supports a 1 million token context window, tool calling, and structured outputs.
The official documentation notes that the provider may retain prompts for model improvement.
8 more timeline events
All Facts (17)
Nemotron 3.5 Lightning is a 30B-scale Mixture of Experts (MoE) model that activates only approximately 3B parameters per inference.
75%Partially VerifiedCline CLI v3.0.74's Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1, and removes Mimo V2.6 Flash and Muse Spark 1.3 Contributor.
75%UnverifiedThe two models share the same training philosophy: large-scale distillation, long-context training, and native speculative decoding.
50%UnverifiedSuper is a 120B-parameter, 12B-activated-parameter multimodal model; Lightning is a 30B-parameter, 3B-activated-parameter text-only model.
50%UnverifiedBoth models use a mixture-of-experts (MoE) and hybrid Mamba-Transformer architecture, targeting reasoning, long-context understanding and agentic workloads.
50%UnverifiedThe video mentioned that the technical details of Nemotron 3.5 Lightning will also be included in the technical report, showing that this generation of the product line has more than one model.
50%UnverifiedNemotron 3.5 Lightning is positioned for long-running agents, sub-agent collaboration, chatbots, and RAG systems, rather than one-off question answering.
50%UnverifiedNemotron 3.5 Lightning uses a hybrid structure that interleaves Mamba-2 with MoE layers, paired with a small number of attention layers.
50%UnverifiedNemotron 3.5 Lightning supports a 1 million token context window, tool calling, and structured outputs.
50%UnverifiedThe official documentation notes that the provider may retain prompts for model improvement.
50%UnverifiedFor teams handling sensitive data, private code, or high compliance requirements, using free models that retain prompts requires evaluating data anonymization or switching to other deployment methods.
50%UnverifiedWhen running the NVIDIA Nemotron 3.5 Lightning model, the Mac mini generates approximately 38 tokens per second, while the Mac Studio generates approximately 124 tokens per second.
50%UnverifiedThe article recommends that beginners start with Nemotron 3.5 Lightning paired with LoRA, because this combination can lower the entry barrier to a range affordable with a single mid-to-low-end GPU.
50%UnverifiedNemotron 3.5 Lightning is reportedly distilled from the larger Nemotron 3 Ultra.
50%UnverifiedSamaya AI released Sterling, a post-trained model for financial research scenarios, built on Nemotron 3.5 Lightning with approximately 3B activated parameters.
50%UnverifiedNVIDIA的Nemotron 3.5 Lightning模型经过2-bit量化后仍能持续运行工具调用长达10分钟以上
50%UnverifiedNVIDIA推出了名为Nemotron 3.5 Lightning的开源模型
50%