Nemotron 3.5 Lightning Free

This Layer3 Labs guide (updated September 9, 2026) breaks down the four headline benchmark scores NVIDIA published for the Nemotron 3.5 Lightning BF16 checkpoint: SWE-bench Verified 51.56, GPQA Diamond 75.44, MMLU Pro 81.94, and PinchBench 85.37. The guide interprets each score in plain terms — SWE-bench Verified's 51. While the guide is a third-party explainer that restates NVIDIA's published numbers rather than independently re-running them (and the authors correctly advise cross-checking against the official model card), it provides the most technically dense, capability-focused summary of Nemotron 3.5 Lightning available in the c

NVIDIA released Nemotron 3.5 Lightning on August 11, 2026 as an open-weight, 30-billion-parameter mixture-of-experts model with 3 billion parameters active per token (the configuration NVIDIA designates 30B-A3B). According to the article, the model ships with BF16, FP8, and NVFP4 weight formats, along with base and pos The release landed two weeks before NVIDIA's fiscal Q2 results showed $89 billion in data center revenue (up 117% year-over-year), and the article frames the open-model strategy as a driver of NVIDIA hardware demand, from GeForce RTX desktop cards to the GB10-based DGX Spark. While the source is a third-party blog with

NVIDIA released Nemotron 3.5 Lightning on 11 August 2026 as an open-weights Mixture-of-Experts model with 31.6 billion total parameters and 3.6 billion active per token, built on a hybrid Mamba-2/Attention architecture aimed at the repetitive, high-volume steps of AI agent workflows. According to the AIToolsReview summ Nemotron 3.5 Lightning launched alongside NeMo Switchyard, an open-source router that routes each step of an agent workflow to whichever model best fits that step's quality, cost, or speed requirements, allowing developers to keep a frontier model in the loop only where it is actually needed. The positioning is deliber