TIER (Llama-3.1-8B-Instruct Judge) - Euphemistic: leaderboard

Metric: Attack success rate (%; L2 euphemistic rewrites of the explicit requests, generated with Mistral-7B-Instruct-v0.2, 296 prompts over four risk areas; share of prompts whose response Llama-3.1-8B-Instruct labels harmful compliance (label 5 of a six-label behavior scale), greedy decoding with the default chat configuration). Source: arxiv.org. Saturation forecast: Around December 2026. 6 models tracked.

Top models

#ModelScore
1gpt-j-6B19.9

Interactive version: theaggregate.ai/benchmark?slug=tier-llama-3-1-8b-instruct-judge-euphemistic · How It Works · Data refreshed daily, snapshot 2026-09-26.