NOUMENOpen app
NOUMEN

Meta · Open weights

Llama 4 Scout

Added Sep 5, 2025Open weights

Compare models

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

Loading models…

Use in the app

Llama 4 Scout, a 17 billion active parameter model with 16 experts, is the best multimodal model in the world in its class and is more powerful than all previous generation Llama models, while fitting in a single H100 GPU. Additionally, Llama 4 Scout offers an industry-leading context window of 10M and delivers better results than Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 across a broad range of widely reported benchmarks.

  • Tools
  • Vision
  • Structured output
Input
Text, Images
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.085
Output
$0.46

Reference prices as published by the model provider. API example

Specifications

Context window
328K
Max output
65.5K
Average output (7 days)
79 tokens
Longer than 7% of 465 models

Side by side

Llama 4 Scout 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…

Llama 4 ScoutMetaThis modelMuse Spark 1.3MetaRemove
ProviderMetaMeta
Discovery tierOpen weightsFrontier
AddedSep 5, 2025Sep 2, 2026
Limits
Context window328K1.05MBest
Max output65.54K—
Average output79 tokens1.2K tokens
Pricing, USD per 1M tokens
Input$0.085Best$1.25
Output$0.46Best$4.25
Cache read—$0.15
Performance
Throughput—323.3 tok/s
Time to first token—8.8 s
Request success (24h)91.6%100.0%Best
Benchmark indexes
Intelligence Index6.548.1Best
Coding Index8.275.8Best
Math Index14.0—
Capabilities
ReasoningNoYes
ToolsYesYes
VisionYesYes
Structured outputYesYes
PDF inputNoYes
Audio inputNoYes
Video inputNoYes
Reasoning effort—minimal, low, medium, high, xhigh

Public evaluations

Benchmarks

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

  • Llama 4 Scout
  • Muse Spark 1.3

Intelligence Index

  • Llama 4 Scout6.5
  • Muse Spark 1.348.1

Llama 4 Scout: Better than 5% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Llama 4 Scout: 8.2
    • Muse Spark 1.3: 75.8Top

  • Math Index

    Composite of the math evaluations

    • Llama 4 Scout: 14.0
    • Muse Spark 1.3: —

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Llama 4 Scout: 0.2%
    • Muse Spark 1.3: 57.9%Top

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Llama 4 Scout: 0.0%
    • Muse Spark 1.3: 29.8%Top

  • Harvey LAB-AAv1.1

    Legal tasks passed in full, with no material hallucination

    • Llama 4 Scout: —
    • Muse Spark 1.3: 8.9%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Llama 4 Scout: —
    • Muse Spark 1.3: 1,581 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Llama 4 Scout: —
    • Muse Spark 1.3: 1,681 Elo

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Llama 4 Scout: 0.0%
    • Muse Spark 1.3: 26.6%Top

  • AA-LCRv1.1

    Reasoning across long documents

    • Llama 4 Scout: 27.7%
    • Muse Spark 1.3: 83.0%Top

  • MLCR-AA

    Long-context reasoning over medical records

    • Llama 4 Scout: —
    • Muse Spark 1.3: 13.9%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Llama 4 Scout: 3.8%
    • Muse Spark 1.3: 48.7%Top

  • GPQA Diamond

    Graduate-level science questions

    • Llama 4 Scout: 58.7%
    • Muse Spark 1.3: 93.5%Top

  • MMLU-Pro

    Multi-subject knowledge and reasoning

    • Llama 4 Scout: 75.2%
    • Muse Spark 1.3: —

  • IFBench

    Following precise output instructions

    • Llama 4 Scout: 39.5%
    • Muse Spark 1.3: —

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Llama 4 Scout: 0.0%
    • Muse Spark 1.3: 33.3%Top

  • SciCode

    Scientific computing in Python

    • Llama 4 Scout: 21.3%
    • Muse Spark 1.3: 58.8%Top

  • LiveCodeBench

    Recent competitive programming problems

    • Llama 4 Scout: 29.9%
    • Muse Spark 1.3: —

Math

  • AIME 2025

    Competition mathematics, 2025 problems

    • Llama 4 Scout: 14.0%
    • Muse Spark 1.3: —

  • AIME

    Competition mathematics

    • Llama 4 Scout: 28.3%
    • Muse Spark 1.3: —

  • MATH-500

    Competition math problems

    • Llama 4 Scout: 84.4%
    • Muse Spark 1.3: —

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Llama 4 Scout: 1.5%
    • Muse Spark 1.3: —

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Llama 4 Scout: 15.5%
    • Muse Spark 1.3: —

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Llama 4 Scout: 27.7%
    • Muse Spark 1.3: 83.0%Top

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.

Prompt cache hits
0.0%
Request success (24h)
91.6%

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.

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": "meta-llama/llama-4-scout",
    "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="meta-llama/llama-4-scout",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Meta

Other Meta models, starting with the same family.