Nvidia launches new open AI model — and promises 30% faster performance
Nvidia has introduced the open-source Nemotron 3.5 Lightning model, designed for complex multi-agent AI tasks. The company claims it delivers 30% faster performance compared to similar models in its category.
Nvidia is increasing its commitment to open models, launching today (Tuesday) Nemotron 3.5 Lightning, designed to perform complex tasks with AI agents.
"The world needs open models, now more than ever," said Kari Briski, Nvidia's vice president of generative AI software products, at a press briefing. "They allow developers to run AI in a better and more efficient way. Efficient agents need a system of models, not just one. A complex task requires a front-end model, while simple tasks require a simpler and more efficient model. Nvidia has been building open Nemotron models for many years; they are at the forefront of efficiency and accuracy."
This message continues a line that Nvidia and its CEO, Jensen Huang, have been promoting recently. In July, Huang called for the promotion of open AI models, arguing that they increase competition, reduce dependence on a single technology provider, and allow companies and governments to download models, adapt them to their needs, and run them on their own infrastructure. Dozens of companies and organizations joined the call, including Meta, Microsoft, IBM, AMD, and Hugging Face.
According to Nvidia, Nemotron 3.5 Lightning combines the capabilities of the advanced Nemotron Ultra model with the speed and efficiency of the smaller Nemotron Nano. The model, which has 30 billion parameters, was developed to work with multi-agent systems. According to the company, it allows AI agents to complete tasks 30% faster compared to similar models in its category.
It is an open model that allows organizations to perform fine-tuning, integration into corporate workflows, and execution in a local computing environment. Leading organizations have already begun using it in their operations, including CrowdStrike in cybersecurity, Harvey in the legal field, CodeRabbit in code review, Lila Sciences in life sciences, and Fastino Labs in software development, finance, and healthcare.
Alongside the new model, Nvidia launched NeMo Switchyard, an open-source library for routing between AI models.
"Organizations can integrate NeMo Switchyard into the work of AI agents to allow the agent to route each task to the model best suited for it, without developers needing to rewrite their applications," the company stated. "Combined with Nemotron 3.5 Lightning, the solution gives organizations and developers greater control over how AI tools are run on personal computers, workstations, data centers, and in the cloud."
