ARC-AGI-3 · Phase 1: ARA world-model full records Collection Full record per harness×model×game: trajectories, recordings, replays + the ARA the agent built live. Same game across models = comparison unit. • 73 items • Updated 1 day ago
FM Open Problems · agent trajectories Collection Foundation-model open-problems trajectories, one repo per model. • 3 items • Updated 1 day ago
LHTB · agent trajectories Collection Live Hard Task Bench trajectories, one repo per model. • 5 items • Updated 1 day ago
DiscoverPhysics · Agent-Native Research Artifacts Collection ARAs from agents solving DiscoverPhysics, one repo per model. • 8 items • Updated 1 day ago