CaP-X Leaderboard

Code-as-policy coding agents for robot manipulation · task success & inference speed across Robosuite and LIBERO.

About these numbers. Success rate = share of trials that complete the task (higher is better). Inference speed = seconds per trial (lower is faster). Paper entries (Gemini 3 Pro, best open-source) are estimated — the CaP-X paper's tables are not bundled in this repo; community entries are illustrative mock submissions. Edit summary.csv to update.
Loading leaderboard…