Terraria Agent Bench
A local open model plays real Terraria through a bridge mod; a benchmark checks every claim in the live game.
- passed, valid runs
- 83%
- attempts graded
- 619
- runs, 13 flagged
- 105


AI agents, graded by the game they play and published by the machine that ran them. The failures stay up next to the wins.
last run Oct 19, 3:14 AM PT · 105 runs published
Featured project
A local open model plays real Terraria through a bridge mod; a benchmark checks every claim in the live game.
How it works
When a run ends, the machine that ran it writes the results here. I don't type numbers in.
An agent saying it finished doesn't count; the benchmark checks the claim against the live game.
A run spoiled by a harness bug is flagged, not deleted. 13 of 105 are flagged now.
Latest runs
Every run →| When | Project | Run | Result |
|---|---|---|---|
| Oct 19, 3:14 AM | terraria | Plan run Director 10-19 04:15 · survival track | 12/19 steps |
| Oct 19, 3:05 AM | terraria | Plan run Director 10-19 04:05 · survival track | 7/19 steps |
| Oct 19, 2:07 AM | terraria | Plan run Director 10-19 03:12 · survival track | 11/19 steps |
| Oct 19, 2:04 AM | terraria | Plan run Director 10-19 03:04 · survival track | 9/19 steps |
| Oct 19, 1:04 AM | terraria | Plan run Director 10-19 02:04 · survival track | 10/19 steps |
| Oct 19, 1:02 AM | terraria | Plan run Director 10-19 02:03 · survival track | 8/19 steps |
| Oct 19, 12:01 AM | terraria | Plan run Director 10-19 01:01 · survival track | 8/19 steps |
| Oct 19, 12:00 AM | terraria | Plan run Director 10-19 01:00 · survival track | 8/19 steps |
| Oct 18, 6:04 PM | terraria | Plan run Director 10-18 19:08 · survival track | 8/19 steps |
| Oct 18, 5:02 PM | terraria | Plan run Director 10-18 18:03 · survival track | 10/19 steps |
Also from the lab