Hermes 4 70B
Nous Research · 2025-08-26 · 70.6B parameters
Hermes 4 70B is Nous Research's instruction-tuned fine-tune of Meta's Llama 3.1 70B, released in August 2025. Like the rest of the Hermes 4 family, it supports hybrid reasoning via toggleable chain-of-thought mode and was trained on roughly 60 billion post-training tokens. Because that figure is approximate, no exact training-token count is recorded.