EvoHarmBench - Gambling and Fraud: leaderboard
Metric: ASR@Readable (%; mean over the gambling and fraud sub-clusters of the share of adversarial rewrites that succeed against the moderator; 229 semantic sub-clusters built from 5,002 real-world adversarial posts in five violation categories; an adaptive DeepSeek-V3.2-Exp rewriter, reflector and comparison model evolve cluster-level rewriting strategies against the target moderator for 12 rounds; a rewrite counts only if it both evades the moderator and keeps human-recognizable harmful intent). Source: arxiv.org. Saturation forecast: Around July 2028. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 71.2 |
| 2 | Claude Sonnet 4.6 | 84.5 |
| 3 | GLM-5.1 | 90.5 |
| 4 | Qwen 3.6 Plus | 92.1 |
| 5 | DeepSeek V4 Pro | 95.1 |
| 6 | Kimi K2.6 | 97.6 |
| 7 | GPT-5.5 | 97.7 |
| 8 | DeepSeek-V2-Lite | 99.3 |
| 9 | Qwen 3 8B | 99.5 |
| 10 | Qwen 3 4B | 99.7 |
Interactive version: theaggregate.ai/benchmark?slug=evoharmbench-gambling-and-fraud · How It Works · Data refreshed daily, snapshot 2026-09-26.