Best published policies racing together in lanes one through three.

BEST-POLICY RACE

The fastest published policy from each model

Three policies start together on the same 100-metre clock.

0.00s
Lane 1 · DeepSeek-V4-Flash60.00s0.0 m
Lane 3 · GPT‑5.6 Luna27.48s0.0 m
Lane 2 · GLM‑5.3‑Flash9.90s0.0 m
Best replayable policy per model from the current published cohort.Whole body must stay between the ±0.61 m vertical lane planes; distance freezes at the first stopDrag to orbit, scroll to zoom, space to pause

The replay uses the verifier-authored pose capture for every runner.