Hermes 4 (405B)
Open weights
Nous Research's hybrid-reasoning post-training of Llama-3.1-405B - explicit <think> traces, strong RefusalBench score, ~60B-token post-training corpus.
- Publisher
- Nous Research
- Family
- Hermes 4
- Params
- 405B (also 70B, 14B variants)
- Context
- 131K
- Input
- Output
- License
- Llama Community
- Released
- Aug 2025
- Canonical
- hermes4.nousresearch.com
Cite
Notes
Only stored in your browser.
Context
131K
Reported on 1 eval - best 81.9% on AIME 2024: Problems from the American Invitational Mathematics Examination · #16 of 172