Nous: Hermes 4 405B

Nous Research

open_source_flagship

Capabilities

Structured OutputExtended ThinkingLong Context

Strengths

Exceptional long-context handling (131K tokens)
Hybrid reasoning mode for deliberative tasks
Fully open-weight for self-hosting and privacy
Strong performance on knowledge-intensive tasks (MMLU 0.92)

Weaknesses

Extremely slow inference (38-39 tok/s)
Weak on low-level reasoning and problem-solving
High deployment cost due to 405B parameters
Limited vision capabilities (text-only)

Pricing

Input / 1M tokens

Output / 1M tokens

Context window

131.072k

Transparency

Open weights

10.0 / 10

Open training data

0.0 / 10

Open methodology

7.0 / 10

Licence openness

6.0 / 10

Provider disclosure

6.5 / 10

FMTI company score

N/A

Composite:5.5 / 10

Open-weight model based on Meta's Llama 3.1 405B. Nous Research discloses hybrid reasoning architecture and training approach. Full weights available for self-hosting. Training data composition not fully disclosed. Licence permits commercial use under Meta's terms.

Sustainability

Inference energy

N/A

Training footprint

N/A

Provider infrastructure

5.0 / 10

Composite:5.0 / 10

Open-weight model enables self-hosting for energy efficiency control. 405B parameter count suggests high inference energy cost. No public provider infrastructure disclosures from Nous Research.

Task Fitness

question answering

50%

code

36%

mathematics

43%

business communication

50%

vision

analysis

44%

extraction

30%

long-form generation

75%

research

50%

reasoning

26%

summarisation

80%

translation

70%

conversation

75%