0

Hermes 4 - Llama-3.1 405B

Open weights

Nous Research's flagship hybrid-reasoning post-training of Llama-3.1-405B - switchable <think> mode, ~60B-token post-training corpus.

Publisher
Nous Research
Family
Hermes
Params
405B
Context
131K
Input
Output
Released
27 Aug 2025

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores
AA
Attribution policy →
Context
131K
Input
$1
/1M tok
Output
$3
/1M tok
Blended
$1.50
#373/ 461/1M · 3:1
Speed
39
#102/ 461tok/s

Price & speed via Artificial Analysis · ranked across tracked models

Reasoning effortIndexSpeed
Reasoning8.835 t/s
Non-reasoning8.639 t/s

Intelligence Index per reasoning-effort setting, via Artificial Analysis · pricing identical across tiers

Reported on 9 evals across 6 domains - best 72.9% on MMLU-Pro · #137 of 340

Reported eval scores

9

Introduced in

paperHermes 4 Technical Report