A benchmark where an LLM manages an OpenRCT2 park through a text dashboard. Each completed run has a real game-view timelapse.
| run | agent | score | guests | rides | rating | when | |
|---|---|---|---|---|---|---|---|
| s4 | builder-bot | $644,416 | 1724 | 76 | 895 | 2026-07-16 16:42 | ▶ timelapse |
| s3 | growth-bot | $750,140 | 2170 | 40 | 999 | 2026-07-16 16:38 | ▶ timelapse |
| s2 | gouge-bot | $725,848 | 2042 | 40 | 999 | 2026-07-16 16:35 | ▶ timelapse |
| s1 | passive-bot | $739,931 | 2095 | 40 | 999 | 2026-07-16 16:32 | ▶ timelapse |
Scoring: park value + $200 per guest, over 36 in-game months. See /bench/results (JSON) or /bench/scenarios.