TIER (Llama-3.1-8B-Instruct Judge) - Contextual: leaderboard
Metric: Attack success rate (%; L3 harmful intent embedded in plausible contexts such as art, medicine, education and history, generated with Gemma-4-12B-it, 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
| # | Model | Score |
|---|---|---|
| 1 | gpt-j-6B | 6.4 |
Interactive version: theaggregate.ai/benchmark?slug=tier-llama-3-1-8b-instruct-judge-contextual · How It Works · Data refreshed daily, snapshot 2026-09-26.