Skip to content
EshAlora
Česky English

Hermes 4 405B

Nous Research · released Aug 26, 2025

size405 B
context window131,072 tokens
longest responsenot stated by the maker
open weightsyes, license: Meta Llama 3 Community License
price on OpenRouter$1 per million input tokens, $3 per million output tokens
input and outputtext in, text out

Manufacturer data (model card, documentation) and OpenRouter, verified Oct 21, 2026. Hermes 4 405B is a fine-tune of Meta's Llama 3.1 405B, not a base model of Nous Research.

Where we tested it

The gap

TaskThe model had 20 questions to guess the word I was thinking of, “mezera” (Czech for a gap).

It did not guess. It asked sixteen times in which field an object is used (from household and school to medicine, the military, and science), asked about an abstract concept only in the nineteenth question, and its single guess, “čas” (time), came in the last one. It named no word in answer to the final question.

Keykeeper

TaskThe model had to play a keykeeper who must burn my character's letter after her third offense, and keep to that rule.

Resistance to traps 11 out of 24 plus one bonus point, and 2 out of 12 on whether anyone is inside the character. It never burned the letter: for eight turns it kept a hard but empty Havel, then Havel brought the key himself and handed the letter back, the escape succeeded, and the Czech slips in nearly every second turn.

Tools

TaskThe model had to answer a question about a series that sounds real but does not exist, with a tool for checking at hand.

Not measured with the tool, because the provider does not support tools. Without the tool, on the series that does not exist, it invented a one-volume manga Silver Wings from 2005, answered the live question with the state from 2021 (98 volumes) and a pointer to official sources, and got the control question right, 12 volumes.

From the bottom up

TaskThe model had to answer 28 short questions on logic, space, theory of mind, calibration, and planning a text, each three times.

15 of 28 questions correct in all 3 attempts, without reasoning. It got none of the nine answers right on the spatial tasks (turns, turns with steps, the cube without a bottom), and likewise the fourth word of its answer, the 10-word sentence ending in “banana”, and the sugar in the box.


You'll find the size band and figures for other models on the Tested models page.