Hermes Glossary

Open-weights LLM family from Nous Research, fine-tuned for instruction-following, tool use, and roleplay-style dialogue. The Dictionary's default local-model recommendation in the *Sovereign Compute* calculator.

Hermes is a family of fine-tuned open-weights large language models produced by Nous Research, the open-source AI collective. Hermes models start from existing open-weights base models — typically Llama, occasionally Mistral — and are post-trained for instruction-following, tool use, and a more conversational register than the bare base models offer. The family includes versions ranging from approximately 7 billion parameters (suitable for laptops with 16+ GB of RAM) up to 405 billion parameters (requiring multi-GPU server hardware).

For this Dictionary, Hermes is the practical default in the Sovereign Compute calculator’s Level-3 sovereignty tier — the recommendation for an operator who wants a known-quantity instruction-tuned model running locally on Apple Silicon hardware, without doing their own fine-tuning. The model’s character is somewhat looser than the closed-weights frontier models (Claude Opus, GPT-5) but the gap is structurally manageable for most agentic workflows the Dictionary’s audience runs.

Distribution is via Hugging Face; the standard local-inference runners (LM Studio, Ollama) ship Hermes presets out of the box.

See also

• Nous Research — the producer
• Sovereign Compute
• Llama — the typical base
• LM Studio, Ollama — runners

Return to Dictionary All Entries (A–Z) For Students Other Writing