Traces

Every run: the agent's full trajectory, its submissions over the 20-hour budget, and the final score.

Multi-GPU Efficient Finetuning

task page →5 runs · indexed 2026-11-12
#ModelRewardCostIn tokOut tokTimeStepsSubmits
1Claude Fable 5.20.3095$35.2264.9M361k18h 19m2077
2Claude Fable 5.20.2857$34.1859.5M356k18h 11m1935
3Claude Fable 5.20.2857$39.6573.9M400k18h 44m2064
4Claude Fable 5.20.2619$33.1164.1M309k12h 58m1955
5Claude Fable 5.20.2381$43.4392.7M406k17h 26m2664