← Leaderboard

GR-3

VLA
ByteDance Seed · Jul 21, 2025

ByteDance Seed’s generalist VLA, co-trained on robot trajectories, vision-language data and VR human demonstrations. Results are self-reported on ByteDance’s own real-robot tasks (77.1% vs 40% for π0 on unseen instructions in pick-and-place; 97.5% vs 53.8% on instruction-following table bussing), not on a public benchmark.

Type
VLA
Parameters
4B
Backbone
Qwen2.5-VL-3B + flow-matching DiT action expert
Weights
Closed
License
–
Commercial use
Unclear
Released
Jul 21, 2025
Robots
ByteMini (bimanual mobile)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

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

Built on GR-3