Llama3 — benchmark results

A Meta Llama 3 model row (April 2024 family, released as 8B and 70B base/instruct under the Llama 3 Community License); board label does not specify size or variant. Provider: Meta. Released 2024-04-18. Access: Open.

Unified ELO 1414 ± 45, rank #1163 of 1776 rated models, from 13 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Fibble2 Arena3.03Win Rate (%)88.9
Open LLM Leaderboard - GPQA8.05Score67.9
Fibble Arena6.25Win Rate (%)54.3
RULER86.5Avg Accuracy (%)52.5
Open LLM Leaderboard - MMLU-Pro24.02Score42.6
Open LLM Leaderboard - BBH26.71Score42
Fibble4 Arena0Win Rate (%)40.5
Fibble3 Arena0Win Rate (%)39.3
Fibble5 Arena0Win Rate (%)36.7
Wordle Arena3.12Win Rate (%)34.4
Open LLM Leaderboard - MuSR7.1Score33.4
Open LLM Leaderboard - MATH Level 55.66Score31.3

Interactive version: theaggregate.ai/model?slug=llama3 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.