Introducing NVIDIA’s Latest Nemotron AI Models
NVIDIA has unveiled three new open-source AI models under the Nemotron 3.5 series: Super VL and Bolt variants, designed specifically for autonomous AI agents. These models vary in computational power and hardware demands, enabling their use across diverse applications. The flagship Nemotron 3.5 Super VL features 120 billion parameters and delivers performance up to five times faster than its closest competitor in the same category.
Key Features of Nemotron 3.5 Models
The Nemotron 3.5 Super VL excels in image analysis, instruction following, and integration with external tools. During its initial training, it processed approximately 26 trillion tokens of both text and multimodal data. Subsequent fine-tuning involved around 3.8 million examples, with reinforcement learning applied across dozens of environments. Organizations can tailor the Super VL model using NVIDIA’s NeMo toolkit and Python programming scripts. It is accessible through platforms such as Hugging Face, OpenRouter, ModelScope, and NVIDIA’s NIM service.
The Nemotron 3.5 Bolt lineup includes two models optimized for different hardware capacities:
- The Nemotron 3.5 Bolt 74B-A9B employs 2-bit quantization and fits within 24 GB of memory, ranking first among open models in its class on the PinchBench leaderboard and second on BrowseComp.
- The Nemotron 3.5 Bolt E4B requires about 4 GB of memory and is tailored for fast conversational AI and physical AI systems running on NVIDIA Jetson devices. While the larger Bolt model is already available, the Bolt E4B variant is scheduled for release on October 16, 2026.
Leading companies like Scale AI have leveraged these models to develop AI agents for financial tasks at Skydance and to assist EY in merger and acquisition deal reviews. AfterQuery has customized Nemotron to analyze credit agreements for Hebbia. Additional projects focus on utilizing AI for cybersecurity and programming automation.
NVIDIA’s rollout of the Nemotron series highlights the increasing demand for robust AI solutions capable of adapting to diverse business requirements.
Source: NVIDIA
Specifically, the capabilities of Nemotron 3.5 Super VL promise to significantly boost automation efficiency across sectors such as finance and cybersecurity. The ongoing refinement and customization of these models open new avenues for organizations aiming to integrate cutting-edge AI technologies into their operations.