Nvidia has unveiled a new artificial intelligence model that outperforms leading providers such as OpenAI and Anthropic.
The model, called Llama-3.1-Nemotron-70B-Instruct, was released on the AI platform Hugging Face and quickly attracted attention with its outstanding results in several performance tests. Nvidia reports top scores in key tests, including 85.0 points in the Arena Hard Benchmark, 57.6 in AlpacaEval 2 LC and 8.98 in the GPT-4 Turbo MT Bench. These scores outperform respected models such as OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet.
“More helpful” than other AIs?
Nvidia has developed a system that is intended to be “more helpful” than previous chatbots. In developing the new model, Nvidia used Meta’s open-source Llama 3.1 model as a basis and refined it using advanced training methods, including reinforcement learning from human feedback (RLHF), which is supposed to lead to more natural and contextually appropriate responses.
For companies and organizations, Nvidia’s model presents an interesting new option. The company offers free access to use and apply the model via its build.nvidia.com platform, including an OpenAI-compatible API interface.
The introduction of Nemotron also highlights a growing trend in the AI landscape towards models that are not only powerful but also customizable. Companies today need AI that can be tailored to their specific needs, be it for handling customer service requests or generating complex reports.