Beta
Classificação/Llama 3.1-405B
Meta

Llama 3.1-405B

Código aberto

por Meta · Lançado em 2024-01-01

49.3
pontuação média
N/A
Preço de entrada
N/A
Preço de saída
N/A
Janela de contexto
text
Tipo

Tested on 15 benchmarks with 49.3% average. Top scores: ARC AI2 (93.7%), HellaSwag (85.6%), TriviaQA (82.7%).

Pontuações de benchmark

BenchmarkCategoriaPontuaçãoBar
ARC AI2knowledge93.7
HellaSwagknowledge85.6
TriviaQAknowledge82.7
MMLUknowledge79.3
Winograndeknowledge78.4
BBHreasoning77.2
PIQAknowledge71.8
MATH level 5math49.8
GPQA diamondknowledge34.5
OpenBookQAknowledge32.3
WeirdMLcoding21.4
OTIS Mock AIME 2024-2025math9.6
SimpleBenchreasoning7.6
Cybenchcoding7.5
The Agent Companyagentic7.4

Modelos similares

AlibabaQwen-7B
Alibaba
49.3
Google DeepMindGemini 2.0 Flash
Google DeepMind
50.0
DeepSeekDeepSeek-R1 (May 2025)
DeepSeek
48.5
U
Baichuan2-13B
unknown
48.4

Meta Llama 3.1 Cronologia

Llama 3.1 70B Instruct
26.1
$0.40/M in131Kctx7 benchmarks
Llama 3.1 8B Instruct
28.7+2.6
$0.02/M in(-0.38)16Kctx(-115K)8 benchmarks
Llama 3.1-405B
49.3+20.6
N/AN/Actx15 benchmarks