Meta logo

Llama 4 Scout

Meta · Released Apr 5, 2025

Open Sourcelong-contextbudgetvisionCompare this model

$0.08

Input price / MTok

#1 of 27

$0.3

Output price / MTok

#2 of 27

10M

Context window

#1 of 28

109

Tokens / sec

#11 of 21

57.2

GPQA Diamond

#19 of 19

69.4

MMMU

#8 of 8

32.8

LiveCodeBench

#8 of 8

Summary

Llama 4 Scout is a open-source language model from Meta, released on Apr 5, 2025.

At $0.08 input and $0.3 output per million tokens, it ranks #1 of 27 for input price, placing it the most affordable in this dataset.

Its context window of 10M tokens ranks #1 of 28 for context depth in this dataset.

It targets long-context, budget, vision workloads.

Throughput is measured at 109 tokens per second (ranked #11 of 21 for throughput).

10M token context; 17B active params in a 109B MoE, single H100 deployable.

Benchmarks

BenchmarkScoreMax%RankSource
GPQA Diamond57.210057.2%#19/19Meta AI (self-reported on llama.com)
MMMU69.410069.4%#8/8Meta AI (self-reported on llama.com)
LiveCodeBench32.810032.8%#8/8Meta AI (self-reported on llama.com)

Similar Models

Compare all 6 models

Models with the most shared benchmark coverage and closest scores to Llama 4 Scout.

Llama 4 Scout
GoogleGemini 2.5 Pro
MetaLlama 4 Maverick
OpenAIo4-mini
AnthropicClaude Sonnet 4.6
OpenAIo3
GPQA Diamond#19 of 19
Graduate-level science reasoning
MMMU#8 of 8
Multimodal understanding across 30 subjects
LiveCodeBench#8 of 8
Live competitive programming

Compare Llama 4 Scout side by side

Pre-filled with 5 similar models

Open comparison

Speed source: https://artificialanalysis.ai/models/llama-4-scout