RuneBench
Games · 2026-02-26
RuneBench evaluates coding agents that play an emulated RuneScape world by writing and executing TypeScript against rs-sdk while consulting extracted game-wiki files. Its headline 30-minute skill score is the mean of ln(1 + peak XP/min) across 16 skills; published sub-tasks also cover peak gold accumulation from four starting conditions.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 7.3 |
| Gemini 4 Argon | 6.9 |
| GPT-6.1 Sol | 6.9 |
| Claude Opus 5.5 | 6.7 |
| GPT-6 Sol | 6.4 |
| Grok 4.6 | 6.3 |
| Claude Fable 5.1 | 6.1 |
| Claude Fable 5 | 6.0 |
| Gemini 3.7 Flash | 6.0 |
| Gemini 3.8 Flash | 6.0 |
| GPT-5.6 Sol | 5.9 |
| GPT-5.6 Terra | 5.9 |
| Claude Opus 5 | 5.7 |
| Grok 4.5 | 5.7 |
| DeepSeek V4.1 Flash | 5.7 |
| GPT-5.5 | 5.7 |
| Grok 4.7 | 5.7 |
| Gemini 3.6 Flash | 5.6 |
| Muse Spark 1.1 | 5.5 |
| Claude Sonnet 5.5 | 5.5 |
| Gemini 3.5 Flash | 5.4 |
| GPT-5.6 Luna | 5.3 |
| Muse Spark 1.2 | 5.2 |
| GLM-5.2 | 5.2 |
| GPT-6 Luna | 5.1 |
| Opus 4.8 | 5.1 |
| GLM-5.3 | 5.0 |
| Muse Wonder | 5.0 |
| Sonnet 5 | 4.8 |
| Opus 4.7 | 4.7 |
| GPT-5.4 | 4.7 |
| Gemini 3 Flash Preview | 4.7 |
| Muse Spark 1.3 | 4.6 |
| Gemini 3.1 Pro Preview | 4.5 |
| Claude Haiku 5.5 | 4.5 |
| Opus 4.6 | 4.4 |
| DeepSeek-V4-Pro | 4.3 |
| GPT-5.3-Codex | 4.3 |
| DeepSeek-V4-Flash-0731 | 4.2 |
| Kimi K2.7 Code | 4.2 |