Models · family
Hermes
2 MLX aliases · Nous Research's instruct line — Hermes 3 8B + Hermes 4 70B.
Nous Research's Hermes line — heavily-tuned instruct models built for tool-calling and agent workflows. The hermes tool envelope is named after this family; Hermes 4 70B is the largest Apache-2 dense model in the registry.
- family
- Hermes
- aliases
- 2
- lines
- 1
- install
- rapid-mlx serve <alias>
- OpenAI base URL
- http://localhost:8000/v1
Hermes 3 / Hermes 4 · 2 aliases
Hermes 3 8B for 16 GB Macs; Hermes 4 70B for 128 GB+ Mac Studio. The hermes tool envelope is named after this family.
parser: hermes
| alias | hf repo | tool parser | reasoning | flags |
|---|---|---|---|---|
| hermes3-8b-4bit | mlx-community/Hermes-3-Llama-3.1-8B-4bit | hermes | — | spec |
| hermes4-70b-4bit | lmstudio-community/Hermes-4-70B-MLX-4bit | hermes | glm4 | spec |
Notes & caveats
- Hermes 4 70B is the recommended pick for a 128 GB+ Mac Studio.
- Hermes 3 8B is a strong, agent-friendly small model on 16 GB Macs.