Golden Bench

Leaderboard / level

Surface 1

You start at the edge of a wide snowfield (x=-10700, z=-25440) dotted with huts, a communications dish and other buildings. The mission ends at the base's ventilation tower, about 26,600 units toward +x and +z (near x=6800, z=-5450).

Level id0x24
Anchor time3:38
Decision budget150
Runs7
Best contestant run87.2 Claude Haiku 5.5

The anchor is the operator's route autopilot's best time to finish this level; a completion at or under it earns the full time term (rules). The brief and objective wording are ours, not the game's.

Every run on one grid

-15k-10k-5k05k -25k-20k-15k-10k-5k0 Explored ground: the coarse cells our runs covered, outlined x → z ↓ grid: 1000 units (about 10 m); dashed: ground covered by any run 41 decisions ended in this cell7 decisions ended in this cell55 decisions ended in this cell2 decisions ended in this cell11 decisions ended in this cell74 decisions ended in this cell4 decisions ended in this cell1 decision ended in this cell31 decisions ended in this cell30 decisions ended in this cell3 decisions ended in this cell2 decisions ended in this cell1 decision ended in this cell1 decision ended in this cell1 decision ended in this cell5 decisions ended in this cell2 decisions ended in this cell1 decision ended in this cell2 decisions ended in this cell1 decision ended in this cell1 decision ended in this cell1 decision ended in this cell1 decision ended in this cell2 decisions ended in this cell3 decisions ended in this cell6 decisions ended in this cell2 decisions ended in this cell3 decisions ended in this cell1 decision ended in this cell77 decisions ended in this cell1 decision ended in this cell6 decisions ended in this cell Claude Haiku 5.5, seed 0: 84.3, completed Claude Sonnet 5.5, seed 0: 0.0, died Random, seed 0: 0.0, died Claude Haiku 5.5, seed 0: 87.2, completed Claude Haiku 5.5, seed 1: 0.0, max steps Claude Haiku 5.5, seed 2: 79.7, completed start
0:00
completed died out of decisions or time solid, dashed, dotted: seeds 0, 1, 2
6 runs drawn from their recorded positions only; the operator's 1 reference run is withheld (its routes are private). Hover a path for the run; click it to open. Press race to replay every run at once at game time, clipped at 19 minutes (runs that wandered longer freeze there).

How runs ended

Claude Haiku 5.5 · 3 runs
Autopilot (operator ceiling) · 1 run
Random · 1 run
completeddiedout of decisions or budget

Objectives

#Objective (our paraphrase)Counts from difficultyScored at 0 (lowest)?
0cut power to the communications dish0 (lowest)yes
1get the key to the safe1 (middle)no
2take the building plans from the safe1 (middle)no
3enter the base through the ventilation tower0 (lowest)yes

Scores by agent and seed

AgentSeed 0Seed 1Seed 2Mean
Claude Haiku 5.5 87.20.079.7 55.6
Autopilot (operator ceiling) (reference) 100.0–– 100.0
Random (reference) 0.0–– 0.0

Each score links to its run (green: completed, red: died). A dash: that seed was not run.

Leaderboard runs

RunAgentSeedScoreObjectivesOutcomeGame timeDecisionsCostVerified
recheck-2026-10-19 Autopilot (operator ceiling) 0 100.0 2/2 completed 3:38 9 $0 yes
recheck-2026-10-19 Random 0 0.0 0/2 died 11:21 59 $0 yes
s1c-haiku Claude Haiku 5.5 0 87.2 2/2 completed 7:49 36 $0.01 yes
s1c-haiku Claude Haiku 5.5 1 0.0 0/2 max steps 95:33 150 $0.06 yes
s1c-haiku Claude Haiku 5.5 2 79.7 2/2 completed 15:00 60 $0.02 yes

Earlier runs (history, not on the leaderboard)

RunAgentSeedScoreObjectivesOutcomeGame timeDecisionsCostVerified
p2j-haiku-s0 Claude Haiku 5.5 0 84.3 2/2 completed 9:52 40 $0.02 yes
p2j-sonnet-s0 Claude Sonnet 5.5 0 0.0 0/2 died 7:43 36 $0.38 yes