According to Cointelegraph, Nvidia has introduced a new artificial intelligence model, Llama-3.1-Nemotron-70B-Instruct, on October 15. This model is claimed to surpass the performance of leading AI systems, including GPT-4o and Claude-3. Nvidia's AI Developer account announced the model's launch on the X.com social media platform, highlighting its status as a top performer on lmarena.AI’s Chatbot Arena.

Llama-3.1-Nemotron-70B-Instruct is a modified version of Meta’s open-source Llama-3.1-70B-Instruct. The 'Nemotron' component signifies Nvidia’s enhancements. Meta’s Llama models are designed as open-source foundations for developers. Nvidia has refined this model using curated datasets, advanced fine-tuning methods, and its state-of-the-art AI hardware, aiming to create a more 'helpful' AI system compared to OpenAI’s ChatGPT and Anthropic’s Claude-3.

Benchmarking AI models involves comparative testing, where different models are given the same tasks, and their performance is evaluated. Nvidia claims that Nemotron significantly outperforms existing state-of-the-art models. Although Nemotron is not listed on the Chatbot Arena leaderboards, Nvidia asserts that it scored an 85 on the automated 'Hard' test, which would place it at the top of this section.

This achievement is notable given that Llama-3.1-70B is Meta’s mid-tier open-source AI model, with a larger version, Llama-3.1-405B, also available. In comparison, GPT-4o is estimated to have been developed with over one trillion parameters.