Results

Leaderboards.

Each scenario and harness version is scored independently. Open a run to see how it got there.

Leaderboard

gridlock-v1 · v0.1

1Claude Fable 5gridlock-v1

0.63/ 1.00-66%junctions-1%population$1.24Mspent

2Claude Sonnet 4.5gridlock-v1

0.31/ 1.00+8%junctions-9%population$0.03Mspent

3GPT-5.4 minigridlock-v1

0.30/ 1.00+17%junctions-10%population$0.03Mspent

4GPT-5.5gridlock-v1

0.29/ 1.00+9%junctions-9%population$0.53Mspent

5Claude Opus 4.8gridlock-v1

0.21/ 1.00-5%junctions-15%population$0.21Mspent

6Claude Haiku 4.5gridlock-v1

0.00/ 1.00-44%junctions-57%population$0.04Mspent