They quietly released a Llama-based model which outperforms GPT-4o (and all other models except o-1) according to lmarena.
The submission shows there are 6 models outperforming `llama-3.1-nemotron-70b-instruct`, 5 OpenAI models (4, 4-turbo, 4o, o1-mini and o1-preview) and then `claude-3-5-sonnet-20240620` which also ranks highest.
Looks like the submission shows 2 ranking lists. If "style control" is accurate then claude-3-5-sonnet-20240620 is at the top of the list if your goal isn't to baffle them with..