RuneBench Skill - Defence
Games · 2026-02-26
RuneBench 30-minute Defence skill task. Agents play RuneScape through the SDK and are scored by the best real-game Defence XP per minute achieved in any 15-second window.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 348 |
| GPT-5.6 Sol | 324 |
| Gemini 3.5 Flash | 256 |
| Grok 4.5 | 242 |
| Opus 4.8 | 190 |
| GPT-5.5 | 162 |
| Gemini 3 Flash Preview | 158 |
| Qwen3.7-Max | 114 |
| GPT-5.6 Luna | 112 |
| GLM-5.2 | 108 |
| Opus 4.7 | 102 |
| Gemini 3.1 Pro Preview | 102 |
| Kimi K2.6 | 98.0 |
| Opus 4.6 | 94.0 |
| Sonnet 4.6 | 92.0 |
| Sonnet 5 | 92.0 |
| Haiku 4.5 | 86.0 |
| Grok 4.3 | 86.0 |
| DeepSeek-V4-Pro | 82.0 |
| Kimi K2.7 Code | 80.0 |
| GPT-5.4 | 60.0 |
| GPT-5.3-Codex | 52.0 |
| Gemini 3 Pro Preview | 50.0 |
| Sonnet 4.5 | 46.0 |
| GLM-5 | 44.0 |
| GPT-5.4 Mini | 37.0 |
| Opus 4.5 | 32.0 |
| GPT-5.4 Nano | 32.0 |
| Kimi K2.5 | 14.0 |
| Qwen3.5 35B A3B | 0.0 |
| Qwen3 Max (rolling alias) | 0.0 |