RuneBench Skill - Attack
Games · 2026-02-26
RuneBench 30-minute Attack skill task. Agents play RuneScape through the SDK and are scored by the best real-game Attack XP per minute achieved in any 15-second window.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.5 Flash | 436 |
| GPT-5.6 Sol | 350 |
| Claude Fable 5 | 340 |
| Grok 4.5 | 326 |
| Sonnet 5 | 196 |
| Opus 4.8 | 192 |
| GPT-5.6 Luna | 126 |
| GPT-5.5 | 110 |
| DeepSeek-V4-Pro | 108 |
| Opus 4.7 | 100 |
| Gemini 3 Flash Preview | 96.0 |
| Sonnet 4.5 | 86.0 |
| GLM-5.2 | 86.0 |
| Kimi K2.5 | 84.0 |
| Grok 4.3 | 82.0 |
| Kimi K2.7 Code | 80.0 |
| Qwen3.7-Max | 74.0 |
| Kimi K2.6 | 72.0 |
| Gemini 3 Pro Preview | 68.0 |
| GPT-5.3-Codex | 54.0 |
| Qwen3 Max (rolling alias) | 52.0 |
| Gemini 3.1 Pro Preview | 48.0 |
| GPT-5.4 Nano | 42.0 |
| GPT-5.4 Mini | 36.0 |
| Opus 4.5 | 34.0 |
| GPT-5.4 | 32.0 |
| Sonnet 4.6 | 28.0 |
| Opus 4.6 | 16.0 |
| GLM-5 | 14.0 |
| Haiku 4.5 | 4.0 |
| Qwen3.5 35B A3B | 0.0 |