Nous Model Family is the set of language models released by 📝Nous Research, led by the Hermes series and joined by the Nomos 1 mathematics model and the NousCoder-14B programming model.
Hermes 3, released in August 2024, was created by fine-tuning 📝LLaMA 3.1 at 8B, 70B and 405B parameters on a dataset of primarily synthetically generated responses. Nous describes it as highly steerable, with long-term context retention and enhanced agentic function-calling. Smaller follow-ups include Hermes-3-Llama-3.2-3B and DeepHermes-3, an 8B preview with deep reasoning and chat modes.
The Hermes 4 family, released August 26, 2025, comprises hybrid-mode reasoning models based on Llama-3.1-405B and 70B, plus a dense 14B variant for local inference. Hermes 4.3, released in December 2025 and based on ByteDance's Seed-OSS-36B-Base, extends context to 512K tokens and nearly matches Hermes 4 70B at half the parameter count.
Nomos 1, released December 9, 2025, specializes Qwen3-30B-A3B-Thinking-2507 for mathematical problem-solving and proof-writing; it was trained with Hillclimb AI and is designed for the Nomos Reasoning Harness. NousCoder-14B, released January 6, 2026, is post-trained on Qwen3-14B for competitive olympiad programming. Psyche, which Nous lists as training infrastructure rather than a model, is its distributed training network; Hermes 4.3 was the first production model post-trained entirely on it.
Nous states its aim to train state-of-the-art 🏷️#open-source models. 📝Hermes Agent shares the Hermes name but is model-agnostic: its README tells users to run any model they want through Nous Portal, OpenRouter, OpenAI or their own endpoint.