Beta
Classificação/Llama 4 Maverick
Meta

Llama 4 Maverick

Código aberto

por Meta · Lançado em 2025-04-05

28.0
pontuação média
$0.15/1M
Preço de entrada
$0.60/1M
Preço de saída
1.0M tokens (~524 books)
Janela de contexto
multimodal
Tipo

Tested on 14 benchmarks with 28.0% average. Top scores: MATH level 5 (73.0%), Lech Mazur Writing (63.7%), GPQA diamond (56.0%).

Pontuações de benchmark

BenchmarkCategoriaPontuaçãoBar
MATH level 5math73.0
Lech Mazur Writingknowledge63.7
GPQA diamondknowledge56.0
GeoBenchknowledge52.0
Fiction.LiveBenchknowledge46.2
WeirdMLcoding24.5
SWE-Bench Verified (Bash Only)coding21.0
OTIS Mock AIME 2024-2025math20.5
Aider polyglotcoding15.6
SimpleBenchreasoning13.2
ARC-AGIreasoning4.4
HLEknowledge0.9
FrontierMath-2025-02-28-Privatemath0.7
ARC-AGI-2reasoning0.1

Modelos similares

Google DeepMindGemma 2 27B
Google DeepMind
28.0
AnthropicClaude 3 Sonnet
Anthropic
28.3
OpenAIGPT-4o (2024-11-20)
OpenAI
28.6
MetaLlama 3.1 8B Instruct
Meta
28.7

Meta Llama 4 Cronologia

Llama 4 Maverick
28.0
$0.15/M in1.0Mctx14 benchmarks
Llama 4 Scout
18.9-9.1
$0.08/M in(-0.07)328Kctx(-721K)8 benchmarks