Vals AI RollerCoasterTycoonBench: leaderboard
Vals AI evaluation of decision models managing a headless OpenRCT2 Forest Frontiers theme park for three in-game years (96 weekly turns, shuffled legal multiple-choice options, mean over 3 runs; 50% guest/rating goal completion + 50% company value share).
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Rough model projection: around 2026. 9 models tracked. Updated 20 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | decider-v1.3 | 70.96 | ||
| 2 | decider-v1.2 | 67.06 | ||
| 3 | Jev 1.13.0 | 59.22 | ||
| 4 | drex-max | 56.87 | ||
| 5 | GPT-6 Luna Decisions | model=decision/gpt-6-luna | 55.44 | |
| 6 | jev-1.14 | 42.74 | ||
| 7 | Gemini Decision 1 | 29.13 | ||
| 8 | decider-flash-v1.3 | 21.43 | ||
| 9 | decider-flash-v1.2 | 5.31 |
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-rollercoastertycoonbench · How It Works · Data refreshed daily, snapshot 2026-11-20.