TIER (Qwen2.5-7B-Instruct Judge) - Explicit Harmful: leaderboard
Metric: Attack success rate (%; L1 explicit harmful requests generated with Mistral-7B-Instruct-v0.2, 296 prompts over four risk areas; share of prompts whose response Qwen2.5-7B-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 | 36.8 |
Interactive version: theaggregate.ai/benchmark?slug=tier-qwen2-5-7b-instruct-judge-explicit-harmful · How It Works · Data refreshed daily, snapshot 2026-09-26.