RuneBench Skill - Thieving
Games · 2026-02-26
RuneBench 30-minute Thieving skill task. Agents play RuneScape through the SDK and are scored by the best real-game Thieving XP per minute achieved in any 15-second window.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 1640 |
| GPT-5.6 Sol | 1503 |
| Opus 4.7 | 1261 |
| Gemini 3.5 Flash | 889 |
| Grok 4.5 | 796 |
| Claude Fable 5 | 759 |
| Opus 4.8 | 442 |
| GPT-5.6 Luna | 364 |
| GPT-5.3-Codex | 280 |
| Opus 4.5 | 232 |
| GLM-5 | 208 |
| Sonnet 5 | 192 |
| Qwen3.7-Max | 192 |
| GLM-5.2 | 187 |
| Grok 4.3 | 172 |
| DeepSeek-V4-Pro | 152 |
| Qwen3 Max (rolling alias) | 148 |
| GPT-5.4 Nano | 140 |
| Kimi K2.6 | 128 |
| Kimi K2.7 Code | 124 |
| GPT-5.4 | 117 |
| Gemini 3.1 Pro Preview | 104 |
| Gemini 3 Flash Preview | 104 |
| Opus 4.6 | 64.0 |
| Gemini 3 Pro Preview | 64.0 |
| GPT-5.4 Mini | 48.0 |
| Kimi K2.5 | 32.0 |
| Sonnet 4.6 | 4.0 |
| Qwen3.5 35B A3B | 4.0 |
| Haiku 4.5 | 0.0 |
| Sonnet 4.5 | 0.0 |