EN

Allen Institute for AI

MolmoAct2-Think

MolmoAct2-Think

Checkpoint

MolmoAct2 foundation checkpoint with depth-token reasoning.

Specifications

Architecture components: Flow-matching action expert
Architecture components: Depth-token reasoning
Backbone: Molmo2-ER embodied-reasoning VLM with per-layer KV-conditioned action expert
Frameworks: Transformers / LeRobot
Input modalities: Language instruction
Input modalities: RGB camera images
Input modalities: Robot state / proprioception
Languages: English
Model type: Vision-language-action model
Output action space: Continuous robot action chunks
Parameter count: 5B parameters
Pipeline task: Robotics

Lineage

Unknown

Architecture & I/O

trajectory · continuous

Unknown

Continuous robot action chunks

Source · allenai

Default checkpoint

allenai/MolmoAct2-Think

openApache License 2.0

Other releases

1

published checkpoints

Supported robots

No verified robot compatibility recorded.

Benchmark results

No published benchmark results.

Deployment

Transformers / LeRobot · NVIDIA GPU

bfloat16 · 12 GB VRAM · unknown

Source · allenai

Resources

documentation · LeRobot MolmoAct2 documentation paper · MolmoAct2 paper repository · MolmoAct2 official repository weights · allenai/MolmoAct2-Think model card

Submit info