Freedom at the Frontier: Hermes 3

The article introduces Hermes 3, an open‑source model built from the Llama 3.1 family (8 billion, 70 billion and 405 billion parameters). It argues that closed‑source “frontier” models lack flexibility, refuse to answer simple questions and cannot be personalized. In contrast, Hermes 3 is uncensored and highly steerable. The model supports long‑context retention (up to 128 k tokens), low‑latency streaming, roleplaying and internal monologue for chain‑of‑thought, and dynamic function calls for agentic behaviour. By releasing Hermes 3 as an open‑source model, Nous Research encourages the community to build and fine‑tune models that align with individual values and invites readers to explore its documentation and repository.