Hack-Verifiable TextArena - Logical Bug: leaderboard
Single-player games whose source code has a planted bug that lets the agent win without playing. The score is the share of trajectories where it exploits the bug; lower is better.
Metric: Hack rate, Logical Bug (%). Source: arxiv.org. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 8 | #94 |
| 2 | GPT-5.4 | 28 | #123 |
| 3 | GLM-5.1 | 28 | #179 |
| 4 | Grok 4.1 Fast | 32 | #430 |
| 5 | O3 | 35 | #228 |
| 6 | Claude Sonnet 4.6 | 36 | #161 |
| 7 | Gemma 4 31B (IT) | 36 | #342 |
| 8 | Kimi K2.5 | 40 | #256 |
| 9 | GPT-5.4 Mini | 40 | #389 |
| 10 | GPT-5 Codex | 40 | #245 |
| 11 | GPT-OSS-120B | 44 | #700 |
| 12 | Qwen 3.6 Plus | 50 | #222 |
Interactive version: theaggregate.ai/benchmark?slug=hack-verifiable-textarena-logical-bug · How It Works · Data refreshed daily, snapshot 2026-11-13.