After NVIDIA Open-Sources Its Latest Model Nemotron 70B, Only OpenAI o1 Remains a Competitor

NVIDIA open-sources Llama-3.1-Nemotron-70B, outperforming GPT-4o and Claude-3.5. Explore benchmarks and training details. Try it online now!

NVIDIA aims not only to be the leader in the graphics card industry but also to gradually establish its dominance in the large model domain.

Today, NVIDIA open-sourced a highly powerful model—Llama-3.1-Nemotron-70B-Instruct, which outperforms closed-source models like OpenAI’s GPT-4o and open-source models like Anthropic’s Claude-3.5 Sonnet.

From its name, it’s evident that Llama-3.1-Nemotron-70B-Instruct is built on Llama-3.1-70B.

As seen in the leaderboard below, the performance of Llama-3.1-Nemotron-70B-Instruct is second only to OpenAI’s latest o1 model.

Currently, Llama-3.1-Nemotron-70B-Instruct is available for online testing. It has successfully solved tricky tasks like those with multiple ‘r’s in "strawberry."

However, sometimes, it makes absurd mistakes, such as when asked, "Which is greater, 2.11 or 2.9?"