--- title: Nemotron Lightning - NVIDIA's Super Fast Agent MoE description: NVIDIA has released Nemotron 3.5 Lightning, a 30B total / 3B active parameter Mixture-of-Experts model designed specifically for the execution layer of... canonical: https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Nemotron Lightning - NVIDIA's Super Fast Agent MoE | daily.dev og:description: NVIDIA has released Nemotron 3.5 Lightning, a 30B total / 3B active parameter Mixture-of-Experts model designed specifically for the execution layer of... og:url: https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq og:image: https://api.daily.dev/og/posts/caPUktfhQ.png og:image:alt: Nemotron Lightning - NVIDIA's Super Fast Agent MoE og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Nemotron Lightning - NVIDIA's Super Fast Agent MoE **[Sam Witteveen](https://daily.dev/sources/samwitteveenai)** · 9 min read · 0 upvotes · 0 comments ## Summary NVIDIA has released Nemotron 3.5 Lightning, a 30B total / 3B active parameter Mixture-of-Experts model designed specifically for the execution layer of long-running AI agents — handling tool calls, validation, retrieval, summarization, and classification rather than high-level reasoning. Key speed advantages come from a hybrid Mamba-transformer architecture, a baked-in multi-token predictor, and speculative decoding via D-Flash and D-Spark (tuned for DGX Spark hardware), delivering claimed 4x throughput versus comparable models and 30–35% faster inference than Qwen 3.6 MoE on Pinch Bench across 10,000 tasks. The model ships with BFloat16 and NVFP4 checkpoints, is released under the Open MDW license (commercial use, distillation, no attribution required), and comes with post-training datasets and training recipes. Real-world fine-tuning case studies include CrowdStrike achieving Nemotron 3 Ultra-level accuracy at 1/5 the cost, and CodeRabbit/Base10 training the model for one epoch in under three hours for roughly $100. Unsloth provides scripts for consumer-hardware fine-tuning. The model is text-only, not suited as a front-level orchestrator (notably weak on prompt injection resistance), and is designed to complement NVIDIA's SwitchYard routing system. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much faster is NVIDIA Nemotron 3.5 Lightning compared to Qwen 3.6 MoE? Nemotron 3.5 Lightning is 30–35% faster than Qwen 3.6 MoE on Pinch Bench across 10,000 tasks, and NVIDIA claims up to 4x throughput versus similar-sized models from the Qwen family. The speed gains come from a hybrid Mamba-transformer architecture, a baked-in multi-token predictor, and speculative decoding via D-Flash and D-Spark. _Teams choosing between MoE inference options for agent workloads track benchmarks like these on daily.dev._ ### What license does NVIDIA Nemotron 3.5 Lightning use and can I use it commercially? Nemotron 3.5 Lightning is released under the Open MDW license, which permits commercial use, distillation of outputs, and requires no attribution. Effectively, you own the fine-tuned model and can use it however you like, including deploying it in production agent pipelines. _Developers evaluating open-weight models for production agent systems find licensing details like these on daily.dev._ ### How much does it cost to fine-tune NVIDIA Nemotron 3.5 Lightning for a custom use case? CodeRabbit and Base10 fine-tuned Nemotron 3.5 Lightning for one epoch in under three hours at a cost of approximately $100, then used the result as a router in a larger system. Unsloth also provides scripts to fine-tune the model on consumer-level hardware for fully local customization. _Engineers fine-tuning open-weight models for specific agent tasks keep up with practical cost data like this on daily.dev._ ## Similar posts on daily.dev - [NVIDIA Nemotron 3.5 Lightning · Ollama Blog](https://daily.dev/posts/nvidia-nemotron-3-5-lightning-ollama-blog-uy1kba1ur) · Ollama · 3 upvotes · 0 comments - [The Fastest Nemotron Yet: Embracing NVIDIA Nemotron 3.5 Lightning in Kilo](https://daily.dev/posts/the-fastest-nemotron-yet-embracing-nvidia-nemotron-3-5-lightning-in-kilo-wcpgec5ll) · Kilo Blog · 4 upvotes · 2 comments - [Run Highly Efficient and Accurate Multi-Agent AI with NVIDIA Nemotron 3 Super Using vLLM](https://daily.dev/posts/run-highly-efficient-and-accurate-multi-agent-ai-with-nvidia-nemotron-3-super-using-vllm-cpvrpiesj) · vLLM · 1 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#deep-learning](https://daily.dev/tags/deep-learning), [#ai-inference](https://daily.dev/tags/ai-inference), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts) [View this post on daily.dev](https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Nemotron Lightning - NVIDIA's Super Fast Agent MoE","url":"https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq"},"datePublished":"2026-08-11T13:45:50.024Z","dateModified":"2026-08-12T01:28:08.154Z","description":"NVIDIA has released Nemotron 3.5 Lightning, a 30B total / 3B active parameter Mixture-of-Experts model designed specifically for the execution layer of...","image":"https://i.ytimg.com/vi/fonbmFSmuRk/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/fonbmFSmuRk/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Sam Witteveen","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Sam Witteveen","logo":"https://media.daily.dev/image/upload/s--gJm-KsgL--/f_auto/v1711727006/logos/samwitteveenai","url":"https://daily.dev/sources/samwitteveenai"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,deep-learning,ai-inference,mixture-of-experts","timeRequired":"PT9M","video":{"@type":"VideoObject","name":"Nemotron Lightning - NVIDIA's Super Fast Agent MoE","description":"NVIDIA has released Nemotron 3.5 Lightning, a 30B total / 3B active parameter Mixture-of-Experts model designed specifically for the execution layer of...","thumbnailUrl":"https://i.ytimg.com/vi/fonbmFSmuRk/sddefault.jpg","uploadDate":"2026-08-11T13:45:50.024Z","duration":"PT9M","url":"https://api.daily.dev/r/caPUktfhQ","embedUrl":"https://www.youtube.com/embed/fonbmFSmuRk"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Sam Witteveen","item":"https://daily.dev/sources/samwitteveenai"},{"@type":"ListItem","position":3,"name":"Nemotron Lightning - NVIDIA's Super Fast Agent MoE"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nemotron-lightning---nvidia-s-super-fast-agent-moe-capuktfhq#faq","mainEntity":[{"@type":"Question","name":"How much faster is NVIDIA Nemotron 3.5 Lightning compared to Qwen 3.6 MoE?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3.5 Lightning is 30–35% faster than Qwen 3.6 MoE on Pinch Bench across 10,000 tasks, and NVIDIA claims up to 4x throughput versus similar-sized models from the Qwen family. The speed gains come from a hybrid Mamba-transformer architecture, a baked-in multi-token predictor, and speculative decoding via D-Flash and D-Spark. Teams choosing between MoE inference options for agent workloads track benchmarks like these on daily.dev."}},{"@type":"Question","name":"What license does NVIDIA Nemotron 3.5 Lightning use and can I use it commercially?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3.5 Lightning is released under the Open MDW license, which permits commercial use, distillation of outputs, and requires no attribution. Effectively, you own the fine-tuned model and can use it however you like, including deploying it in production agent pipelines. Developers evaluating open-weight models for production agent systems find licensing details like these on daily.dev."}},{"@type":"Question","name":"How much does it cost to fine-tune NVIDIA Nemotron 3.5 Lightning for a custom use case?","acceptedAnswer":{"@type":"Answer","text":"CodeRabbit and Base10 fine-tuned Nemotron 3.5 Lightning for one epoch in under three hours at a cost of approximately $100, then used the result as a router in a larger system. Unsloth also provides scripts to fine-tune the model on consumer-level hardware for fully local customization. Engineers fine-tuning open-weight models for specific agent tasks keep up with practical cost data like this on daily.dev."}}]} ```