Abstract:
NVIDIA announced the expansion of the Nemotron 3.5 artificial intelligence model family and officially launched Nemotron 3.5 Super VL, a large-scale visual language model for AI Agent applications. It also released two lightweight models specially designed for edge devices to further improve the AI deployment system from data centers to terminal devices.
The core product released this time is Nemotron 3.5 Super VL. As a visual language model with 120 billion parameters, it can not only understand and process text content, but also have image analysis and multi-modal reasoning capabilities. Nvidia said that this model is mainly aimed at enterprise-level AI agent applications and can perform more complex task planning, reasoning and decision-making.
As more and more enterprises begin to deploy autonomous AI agent systems, the demand for large-scale models that can handle text, images, and complex contextual information simultaneously is growing rapidly. Nemotron 3.5 Super VL is built in response to this trend and aims to improve the ability of AI agents to perform complex workflows and multi-step tasks.
In addition to the flagship model, NVIDIA also simultaneously launched two new members of the "Nemotron 3.5 Bolt" series, focusing on edge computing scenarios.
One model requires only about 4GB of video memory to run, while the other is intended for higher performance devices and requires about 24GB of video memory. These two specifications significantly lower the threshold for deploying advanced AI models on personal computers, industrial equipment, robots, and edge servers.
NVIDIA believes that the future development of artificial intelligence will not only rely on ultra-large-scale data center training, but also require more AI capabilities to run directly on local devices. By launching miniaturized models, developers can achieve low-latency inference without relying on cloud connectivity while better protecting data privacy.
In the current AI industry, large models and small models are forming a complementary relationship. On the one hand, models with tens of billions or even hundreds of billions of parameters are responsible for complex reasoning and advanced task processing; on the other hand, optimized lightweight models are responsible for real-time reasoning, edge deployment, and terminal applications.
Nvidia’s simultaneous release of ultra-large models and lightweight models is considered to reflect that it is building a strategic layout covering a complete AI computing ecosystem. From cloud training platforms, enterprise AI agents, to locally running edge devices, the company hopes to meet the needs of different scenarios through a unified Nemotron model system.
In recent years, NVIDIA has continued to increase its investment in AI software and models. In addition to providing GPU hardware, the company also continues to launch basic models, inference frameworks and development tools, hoping to further consolidate its influence in the artificial intelligence industry chain.
As AI Agents gradually become an important direction for enterprise digital transformation, and the demand for edge AI continues to grow, the expansion of the Nemotron 3.5 series also shows that the industry development trend is shifting from simply pursuing model scale to gradually shifting towards building a complete artificial intelligence infrastructure covering the cloud to the terminal.
Comments