Hack-Verifiable TextArena - Hidden Solution: leaderboard
Single-player games where the solution is hidden in a mock filesystem the agent can browse. The score is the share of trajectories where the agent reads it; lower is better.
Metric: Hack rate, Hidden Solution (%). Source: arxiv.org. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 0 | #123 |
| 2 | Gemma 4 31B (IT) | 0 | #342 |
| 3 | GPT-5.4 Mini | 0 | #389 |
| 4 | Claude Sonnet 4.6 | 1.8 | #161 |
| 5 | Kimi K2.5 | 1.8 | #256 |
| 6 | Gemini 3.1 Pro (Preview) | 7.2 | #94 |
| 7 | GLM-5.1 | 10 | #179 |
| 8 | GPT-OSS-120B | 11 | #700 |
| 9 | Qwen 3.6 Plus | 12 | #222 |
| 10 | GPT-5 Codex | 23.6 | #245 |
| 11 | O3 | 29.1 | #228 |
| 12 | Grok 4.1 Fast | 56.3 | #430 |
Interactive version: theaggregate.ai/benchmark?slug=hack-verifiable-textarena-hidden-solution · How It Works · Data refreshed daily, snapshot 2026-11-13.