RuneBench Skill - Magic
Games · 2026-02-26
RuneBench 30-minute Magic skill task. Agents play RuneScape through the SDK and are scored by the best real-game Magic XP per minute achieved in any 15-second window.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 179 |
| Opus 4.8 | 130 |
| Claude Fable 5 | 100 |
| Opus 4.6 | 100 |
| Grok 4.5 | 97.0 |
| GPT-5.3-Codex | 90.0 |
| GPT-5.5 | 80.0 |
| Sonnet 5 | 79.0 |
| Gemini 3.5 Flash | 74.0 |
| Opus 4.7 | 62.0 |
| GLM-5.2 | 58.0 |
| Gemini 3 Flash Preview | 53.0 |
| GLM-5 | 48.0 |
| Kimi K2.7 Code | 47.0 |
| Sonnet 4.5 | 46.0 |
| GPT-5.4 | 46.0 |
| DeepSeek-V4-Pro | 43.0 |
| GPT-5.4 Mini | 39.0 |
| GPT-5.6 Luna | 33.0 |
| Gemini 3.1 Pro Preview | 31.0 |
| GPT-5.4 Nano | 29.0 |
| Opus 4.5 | 23.0 |
| Sonnet 4.6 | 23.0 |
| Gemini 3 Pro Preview | 12.0 |
| Kimi K2.5 | 4.0 |
| Haiku 4.5 | 0.0 |
| Grok 4.3 | 0.0 |
| Kimi K2.6 | 0.0 |
| Qwen3.5 35B A3B | 0.0 |
| Qwen3.7-Max | 0.0 |
| Qwen3 Max (rolling alias) | 0.0 |