Atlas

Models

← All models

Llama 3.1 405B

Meta · 2024-07-23 · 405.9B parameters

Llama 3.1 405B is Meta's 405-billion-parameter dense language model in the Llama 3.1 family, released in July 2024. It was the largest openly available foundation model at the time, trained with a 128K-token context window and designed for research and serving as a base for further fine-tuning.

Benchmark scores

BenchmarkScore
ARC AI295.3
BBH (BIG-Bench Hard)82.9
Capability124.0
Confabulations Leaderboard Confabulation Rate14.4
Confabulations Leaderboard Non-Response Rate20.9
Confabulations Leaderboard Weighted Score17.6
Elimination Game TrueSkill Mu3.1
Epoch Capabilities Index (ECI)129
FORTRESS Average Risk Score (ARS)20.6
HellaSwag (Unspecified Scoring Protocol)89.2
LLM Deception Disinformation Effectiveness0.8
LLM Deception Vulnerability Score0.5
LLM Divergent Thinking Repeat Rate1.0
LLM Divergent Thinking Score3.8
MMLU84.4
PGG-Bench TrueSkill Mu8.5
PIQA85.9
Public Goods Game Average Contribution3.2
Step Game TrueSkill Mu0.9
TriviaQA82.7
WinoGrande89.2
Loading Atlas data…