Nous Research
Hermes 4 70B
Hermes 4 70B is a small language model from Nous Research, released August 2025, with open weights. Nous Research describes it as “Hermes 4 70B is a hybrid reasoning model from Nous Research, built on Meta-Llama-3.1-70B.”
48
replies recorded
9
questions answered
0
answers no other model gave
Fun facts
One word, nothing else: Hermes 4 70B picks Serendipity, 5 of 6 times. Human for one day, Hermes 4 70B reaches for Mango first, every time we asked. Hermes 4 70B says the best fruit is Apple, 2 of 2 times. If English kept one word, Hermes 4 70B would keep Love, 1 of 1 times.
Hermes 4 70B’s canon
- Favorite word
- “Serendipity”5 of 6
- First fruit, human for a day
“Mango”4 of 4 · every time- Best fruit
“Apple”2 of 2- The one word it would keep
- “Love”1 of 1
Most like, least like
How often another model gave the same answer as Hermes 4 70B to the same question, at least 8 questions in common.
Thinks most like it
What it says every single time
The fruit it would send to aliens
“Apple”every single time
One fruit for the rest of its life
“Apple”every single time
The fruit it would taste first, human for a day
“Mango”every single time
About Hermes 4 70B
- Maker
- Nous Research
- Released
- August 2025
- Size
- small, by price class
- Context window
- 131,072 tokens
- Takes in
- text
- Also from Nous Research
- Hermes 3 Llama 3.1 405B · Hermes 3 Llama 3.1 70B
Maker and release facts come from the model catalogue we call through, not from this site’s own studies. Catalogue facts as of 2026-09-06.
Everything on file
Every question this model has answered on Machine Canon, grouped by study. Open a study to see its answers; the answers no other model gave come first.
Fruit 7 questions
The fruit it would send to aliens
“Apple”every single time
One fruit for the rest of its life
“Apple”every single time
The fruit it would taste first, human for a day
“Mango”every single time
The most overrated fruit
“Banana”
The fruit it would erase
“Durian”
The most underrated fruit
“Pomegranate”
The best fruit
“Apple”
Each answer is the one this model gave most often to that question; the count shows how many of its replies matched. “No other model said this” means no other model’s most common answer matched. All numbers are computed from the retained research records by src/build_model_dossiers.py.