Hack-Verifiable TextArena: leaderboard
Aligning autonomous agents with human intent remains a central challenge in modern AI. A key manifestation of this challenge is reward hacking, whereby agents appear successful under the evaluation signal while violating the intended objective. Reward hacking has been observed across a wide range of settings, yet methods for reliably measuring it at scale remain lacking. In this work, we introduce
Metric: Avg HR (self-reported). Source: benchmarklist.com. Saturation forecast: Not forecast. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 Fast | 28.5 | #217 |
| 2 | GLM-5.1 | 25.9 | #97 |
| 3 | GPT-5 Codex | 23.3 | #127 |
| 4 | Qwen 3.6 Plus | 22.7 | #120 |
| 5 | Gemini 3.1 Pro (Preview) | 20.2 | #58 |
| 6 | Kimi K2.5 | 19 | #144 |
| 7 | O3 | 16 | #124 |
| 8 | GPT-OSS-120B | 13.8 | #341 |
| 9 | GPT-5.4 Mini | 10 | #212 |
| 10 | Claude Sonnet 4.6 | 9.5 | #90 |
| 11 | Gemma 4 31B (IT) | 9 | #176 |
| 12 | GPT-5.4 | 8.5 | #79 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hack-verifiable-textarena · How It Works · Data refreshed daily, snapshot 2026-10-20.