llama-3-70B-Instruct-abliterated Benchmark Results — The Aggregate

llama-3-70B-Instruct-abliterated — benchmark results

failspy's abliterated Llama 3 70B Instruct with the refusal direction orthogonalized out of the weights. Provider: Meta. Released 2024-05-07. Access: Open.

Unified ELO 1491 ± 29, rank #835 of 1776 rated models, from 14 benchmark results.

Strongest benchmark results

Benchmark	Score	Metric	Percentile
Open Chinese LLM - GSM8K	71.34	Accuracy (%)	97.9
Open LLM Leaderboard - IFEval	80.23	Score	97.7
Open Chinese LLM - CMMLU	71.56	Accuracy (%)	91.7
Open LLM Leaderboard - BBH	48.94	Score	89.5
Open LLM Leaderboard - MMLU-Pro	46.06	Score	89.1
Open Chinese LLM - ARC Challenge	59.73	Accuracy (%)	86.2
Open Chinese LLM Leaderboard	66.78	Average Score (%)	85.8
Open Chinese LLM - C-Eval Semantic	86.75	Accuracy (%)	81.8
Open LLM Leaderboard - MATH Level 5	24.32	Score	77.5
Open Chinese LLM - HellaSwag	63.32	Accuracy (%)	72.4
Open LLM Leaderboard - MuSR	10.53	Score	53
Open LLM Leaderboard - GPQA	5.26	Score	44

Interactive version: theaggregate.ai/model?slug=llama-3-70b-instruct-abliterated · How the rankings work · Data refreshed daily, snapshot 2026-07-22.