← Leaderboard

MolmoAct

VLA
Ai2 · Aug 11, 2025

Ai2’s first action reasoning model: it predicts depth perception tokens, then a 2D visual trace, then actions, and the trace can be edited to steer the robot. It was released with the MolmoAct Dataset. The LIBERO number is self-reported for MolmoAct-7B-D after LoRA post-training.

Type
VLA
Parameters
7B
Backbone
Molmo (Qwen2.5-7B + SigLIP2) for 7B-D; OLMo2-7B for 7B-O
Weights
Open
License
Apache-2.0
Commercial use
Unclear
Released
Aug 11, 2025
Robots
Google Robot (sim), Franka Panda (sim), Franka, Bimanual Franka
SO-101
–
LeRobot
–
Inputs
RGB · L
VRAM
–

Benchmark results

1 of 13 benchmarks
LIBEROSaturatedSimulation
#60 of 7086.8%, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error