Detailed Settings Needed for Reproduction

[MrGGLS]

I'm trying to reproduce the leaderboard scores using OpenClaw (3.13), but it's been difficult due to the lack of specific settings (e.g., timeout, max_steps, context window, etc.). Could you please share the detailed evaluation configuration? Thank you very much!

[RaymondTang2003]

Thanks for your attention. The timeout is set to 2000 seconds, there is no limit on max_steps, and the context window is determined by the inherent context window capacity of the adopted model.

[MrGGLS]

Thanks for the reply. Are you planning to share the detailed evaluation settings/configurations in the repo?