Long-WAM

Long-context world-action model from NVIDIA with MIT, HKU and UCSD that predicts future video latents and then actions from up to 19.2 s of camera history. It also reports LIBERO, RoboTwin 2.0 and RoboCasa365 results, which are not listed: its LIBERO average does not match its per-suite cells, and its RoboTwin training data is not stated.

NVIDIA · Oct 7, 2026 · World-actionReport an error
0Benchmarks reported
Oct 7, 2026Released
OpenWeights
World-actionType

Benchmark results0

No results on the tracked benchmarks yet.

Not reported on 15 benchmarks
Wrong or missing number? Report an error