RuneBench: leaderboard
AI coding agent benchmark where models write TypeScript snippets to play RuneScape via an MCP server, scored by mean ln(1 + peak XP/min) across 16 skills in 30-minute sessions.
Metric: Mean ln(1 + peak XP/min) across sixteen 30-minute skills; higher is better. Source: maxbittker.github.io. Saturation forecast: Around November 2027. 92 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (High) | 7.26 | #6 (GPT-6 Astra) |
| 2 | GPT-6 Astra | 6.91 | #6 |
| 3 | Gemini 4 Argon | 6.91 | #8 |
| 4 | GPT-6.1 Sol (High) | 6.88 | #7 (GPT-6.1 Sol) |
| 5 | GPT-6.1 Sol | 6.79 | #7 |
| 6 | GPT-6.1 Sol (Low) | 6.75 | #7 (GPT-6.1 Sol) |
| 7 | Claude Opus 5.5 (xHigh) | 6.66 | #3 (Claude Opus 5.5) |
| 8 | Claude Opus 5.5 | 6.63 | #3 |
| 9 | GPT-6 Sol (High) | 6.44 | #14 (GPT-6 Sol) |
| 10 | GPT-6 Sol | 6.42 | #14 |
| 11 | Grok 4.6 | 6.28 | #26 |
| 12 | Claude Fable 5.1 (xHigh) | 6.14 | #9 (Claude Fable 5.1) |
| 13 | Claude Fable 5 | 6.01 | #16 |
| 14 | Claude Fable 5.1 | 6 | #9 |
| 15 | Gemini 3.7 Flash | 5.96 | #23 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=runebench · How It Works · Data refreshed daily, snapshot 2026-10-20.