Nvidia announced three new open artificial intelligence models aimed at creating AI agents for businesses. On the same day, Meta unveiled an open-weights model and announced a partnership with Apple to run it locally on high-performance Mac computers.
The launches expand the range of AI options developed in the United States, although it remains unclear whether they will attract users to cheaper Chinese models.
Experts point out that customizing open models can surpass the efficiency of large closed proprietary systems. – Image created by artificial intelligence: Gemini/Olhar Digital
Nvidia bets on AI agents
The three models are part of the Nemotron family. One of them, Nemotron 3.5 Super VL, was developed for tasks that require AI agents to operate continuously.According to Nvidia, the system is faster at long-running activities and processes text, images, and videos.
Kari Briski, Nvidia’s vice president of generative AI software for enterprises, said the company plans to continue releasing enhanced open models.
“This is a message to developers, our partners and businesses that they can count on us,” Briski said in a statement.
According to the executive, interest in Nemotron models has increased because they are developed in the United States and allow the underlying data to be inspected. Government agencies and financial institutions have expressed particular concern about the use of Chinese models.
Companies customize open models
Atlassian, CrowdStrike, Dassault Systèmes, Salesforce and Synopsys are among the companies that have adapted Nemotron with their own data, according to The Wall Street Journal.
The process, known as fine-tuning, consists of adapting an existing model to improve its performance on specific tasks. According to Briski, this customization can produce better results than using state-of-the-art models, precisely by meeting each company’s needs.
The approach also enables:
- Directing the model toward specific activities in each sector.
- Incorporating proprietary information into the adaptation process.
- Modifying an existing system without having to develop another from scratch.
Running artificial intelligence locally on computers, as proposed by Meta and Apple, ensures greater privacy and speed in processing sensitive data. Image: mundissima/Shutterstock
Meta and Apple bet on local AI
Meta announced a new open-weights model that can run locally on high-performance Mac computers in collaboration with Apple. In this format, processing takes place on the device itself, without relying entirely on remote servers.
The initiatives expand the range of American models available to companies and developers. However, there is not yet enough information to determine whether the releases will be able to compete with lower-cost Chinese alternatives.
Valdir Antonelli is a journalist specializing in digital marketing and consumer affairs.
