Nous Research releases Hermes 3

Hermes 3 adapted Llama 3.1 into an open family focused on instruction following, roleplay, tool use, and agentic work.

Nous Research published Hermes 3 across 8B, 70B, and 405B model sizes, built on Llama 3.1. The technical report describes a large instruction dataset that included coding, mathematics, roleplay, tool use, agents, and retrieval.

I like keeping releases like this beside the frontier lab announcements. Open models often show where the broader community wants more control, not only more raw capability.