RevengeBench: leaderboard
Agents reverse-engineer and counter opponent bots across five game arenas (Battlesnake, Halite, poker, Robocode, Robot Rumble); mean score across arenas.
Metric: Arena Mean Score. Source: revengebench.com. Saturation forecast: Around June 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 71.92 | #98 |
| 2 | GPT-5.5 | 71.1 | #31 |
| 3 | GLM-5.1 | 67.44 | #97 |
| 4 | Kimi K2.6 | 59.9 | #104 |
| 5 | DeepSeek V4 Pro | 58.9 | #102 |
| 6 | Gemma 4 31B | 52.28 | #200 |
| 7 | GPT-5.4 Mini | 49.88 | #212 |
| 8 | Grok 4.1 Fast | 47.16 | #217 |
| 9 | Gemma 4 26B A4B | 46.74 | #358 |
| 10 | DeepSeek V4 Flash | 46.14 | #137 |
| 11 | Qwen 3.6 35B A3B | 42.82 | #226 |
| 12 | GPT-OSS-120B | 33.74 | #341 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=revengebench · How It Works · Data refreshed daily, snapshot 2026-10-20.