Hermes 2 is a line of open-weight fine-tunes from Nous Research, an independent AI research group known for taking strong base models and retraining them on curated instruction and conversation data rather than building foundation models from scratch. The line includes versions built on top of Mistral 7B, Mixtral 8x7B, and a 34B-parameter variant based on Yi-34B, with the largest carrying a 4,096-token context window inherited from its base model. Nous trained Hermes 2 on a large set of synthetic and human-curated conversations aimed at general assistant behavior, function calling, and reduced refusals compared to more heavily filtered commercial chatbots. The models are released under Apache 2.0 and distributed through Hugging Face, where they became popular with the open-source LLM community for local deployment via tools like llama.cpp and Ollama. Hermes 2 predates Nous Research’s later Hermes 3 line, which extended the same fine-tuning approach to larger and more capable base models like Llama 3.1.