← Leaderboard

GE-Act 2.0

World-action
AGIBOT · Sep 4, 2026

Second-generation GE-Act, a world-action model whose generative and action parts are trained from scratch on manipulation data: a control-oriented autoencoder, a single-step visual planner and an inverse dynamics model, co-trained on up to 30,000 hours of real-robot data. Results are self-reported on AGIBOT’s own 100-task, 20-skill evaluation, where scaling co-training data from 300 to 30,000 hours raised zero-shot success from 17.1% to 44.1% on G1. Code is listed as coming soon.

Type
World-action
Parameters
—
Backbone
–
Weights
Unknown
License
–
Commercial use
Unclear
Released
Sep 4, 2026
Robots
AgiBot G1 (G1-OP), G2-90D
SO-101
–
LeRobot
–
Inputs
RGB · P · L
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

Not reported on 13 benchmarks
Wrong or missing number? Report an error