Published 13 hours ago • loading... • Updated 5 hours ago
NVIDIA's latest 120B-parameter model completes tasks 5x faster
On Thursday, Nvidia announced the Nemotron 3.5 family, including the 120B Super VL vision-language model and two smaller Bolt edge models for local autonomous agentic tasks.
Targeting specialized agentic work, Super VL delivers "the fastest task completion and highest accuracy for specialized, always-on agents," according to Nvidia, claiming fivefold speed gains over competitors.
Benchmark testing shows Super VL achieving 80.8% accuracy on CharXiv, outperforming competitors in scientific reasoning and terminal tasks, though trailing on SWE-bench Verified at 71.1.
The Bolt 74B-A9B uses 2-bit quantization to operate within 24GB of memory, enabling local agentic tasks, while the smaller Bolt E4B arrives later this month.
Developers can access the models on Hugging Face under the OpenMDW 1.1 license, which permits commercial use and modification for organizations deploying customizable AI agents locally.