NOUMENOpen app
NOUMEN

nousresearch · Open weights

Hermes 4 Large (Thinking)

Added Jan 1, 2024Open weights

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

Hermes 4 Large with thinking enabled. Streams visible reasoning before the final answer and supports structured outputs.

  • Structured output
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.30
Output
$1.20

Reference prices as published by the model provider. API example

Specifications

Context window
128K
Max output
8.2K
Average output (7 days)
1K tokens
Longer than 70% of 468 models

Side by side

Hermes 4 Large (Thinking) compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Hermes 4 Large (Thinking)nousresearchThis modelNvidia Nemotron Super 49BnvidiaRemove
Providernousresearchnvidia
Discovery tierOpen weightsOpen weights
AddedJan 1, 2024Aug 8, 2025
Limits
Context window128K128K
Max output8.19K16.38KBest
Average output1K tokens—
Pricing, USD per 1M tokens
Input$0.30$0.15Best
Output$1.20$0.15Best
Performance
Throughput45.0 tok/s—
Time to first token0.9 s—
Request success (24h)100.0%—
Benchmark indexes
Intelligence Index7.5Best7.3
Math Index69.7Best7.7
Capabilities
Structured outputYesNo
Reasoning effort——

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

  • Hermes 4 Large (Thinking)
  • Nvidia Nemotron Super 49B

Intelligence Index

  • Hermes 4 Large (Thinking)7.5
  • Nvidia Nemotron Super 49B7.3

Hermes 4 Large (Thinking): Better than 11% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Math Index

    Composite of the math evaluations

    • Hermes 4 Large (Thinking): 69.7Top
    • Nvidia Nemotron Super 49B: 7.7

Document reasoning

  • AA-LCRv1.1

    Reasoning across long documents

    • Hermes 4 Large (Thinking): 22.3%Top
    • Nvidia Nemotron Super 49B: 12.0%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Hermes 4 Large (Thinking): 10.9%Top
    • Nvidia Nemotron Super 49B: 3.8%

  • GPQA Diamond

    Graduate-level science questions

    • Hermes 4 Large (Thinking): 72.7%Top
    • Nvidia Nemotron Super 49B: 51.7%

  • MMLU-Pro

    Multi-subject knowledge and reasoning

    • Hermes 4 Large (Thinking): 82.9%Top
    • Nvidia Nemotron Super 49B: 69.8%

  • IFBench

    Following precise output instructions

    • Hermes 4 Large (Thinking): 32.7%
    • Nvidia Nemotron Super 49B: 39.5%Top

Coding

  • LiveCodeBench

    Recent competitive programming problems

    • Hermes 4 Large (Thinking): 68.6%Top
    • Nvidia Nemotron Super 49B: 28.0%

Math

  • AIME 2025

    Competition mathematics, 2025 problems

    • Hermes 4 Large (Thinking): 69.7%Top
    • Nvidia Nemotron Super 49B: 7.7%

  • AIME

    Competition mathematics

    • Hermes 4 Large (Thinking): —
    • Nvidia Nemotron Super 49B: 19.3%

  • MATH-500

    Competition math problems

    • Hermes 4 Large (Thinking): —
    • Nvidia Nemotron Super 49B: 77.5%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Hermes 4 Large (Thinking): 11.4%Top
    • Nvidia Nemotron Super 49B: 0.0%

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Hermes 4 Large (Thinking): 22.2%
    • Nvidia Nemotron Super 49B: —

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Hermes 4 Large (Thinking): 22.3%Top
    • Nvidia Nemotron Super 49B: 12.0%

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Throughput
45.0 tok/s
Faster than 33% of 274 models
Time to first token
0.9 s
Quicker than 98% of 274 models
Prompt cache hits
66.8%
Request success (24h)
100.0%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 100.0%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nousresearch/hermes-4-405b:thinking",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="nousresearch/hermes-4-405b:thinking",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

nousresearch

Other nousresearch models, starting with the same family.