Skip to content
embodied.guide

Back to the AI models

Ai2United States2026

MolmoAct 2

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

Robotsnot stated

Key figures as stated by the developer

5 of 6 stated · 1 gap
  • Model type ⁠A vision-language-action
  • Controls ⁠B arms and hands
  • Inputs ⁠A camera language robot state
  • Access ⁠B open download
  • Parameters not stated
  • Base model ⁠B Molmo 2-ER
How to read the figures
14kgA30.9lbA
The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb
Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (5)

Model type

Developer wording: "MolmoAct2 is an open vision-language-action model for robot control." The blog also calls it an action reasoning model, a term with no value in the closed list: "the first Action Reasoning Model (ARM)".

Controls

"a dedicated action expert that generates robot actions through flow matching". "MolmoAct2 supports out-of-the-box deployment on three robot embodiments:" SO-100, bimanual YAMs and Franka DROID, all arms.

Inputs

Model card: "MolmoAct2 generate robot actions from visual observations and language instructions". Code repository: the model "adds robot state and action modeling".

Access

"are available for researchers to study and build on—including the model weights".

Base model

"MolmoAct 2 isn’t simply initialized from Molmo 2, but Molmo 2-ER".

Full specifications

8 of 12 fields stated
The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field Value Source and caveats
Input and output
Model typevision-language-actionA

Developer wording: "MolmoAct2 is an open vision-language-action model for robot control." The blog also calls it an action reasoning model, a term with no value in the closed list: "the first Action Reasoning Model (ARM)".

Inputscamera language robot stateA

Model card: "MolmoAct2 generate robot actions from visual observations and language instructions". Code repository: the model "adds robot state and action modeling".

Controlsarms and handsB

"a dedicated action expert that generates robot actions through flow matching". "MolmoAct2 supports out-of-the-box deployment on three robot embodiments:" SO-100, bimanual YAMs and Franka DROID, all arms.

Degrees of freedom controllednot stated
Size and speed
Parametersnot stated
Control ratenot stated

Note The model has no single control rate; it "predicts fixed-horizon action chunks (30 steps at 30 Hz for YAM/SO-100/101, 15 at 15 Hz for DROID, 10 at 10 Hz for LIBERO) and executes them open-loop". The inference figure ("Enabling CUDA Graphs gives the larger gain: MolmoAct2 reaches 55.79 Hz") "is amortized chunk throughput, not closed-loop reactivity". ⁠C
Where it runs and what it is built on
Runs wherenot stated
Base modelMolmo 2-ERB

"MolmoAct 2 isn’t simply initialized from Molmo 2, but Molmo 2-ER".
Openness
Accessopen downloadB

"are available for researchers to study and build on—including the model weights".

LicenseApache 2.0D

Code repository, section License: "This model is licensed under Apache 2.0." The repository adds that the model is meant for research and education under Ai2's Responsible Use Guidelines.
The developer’s own claims
Learns new tasksneeds retrainingC

Developer claim: "MolmoAct2 can be adapted to novel tasks and unseen embodiments via fine-tuning". The model card calls the checkpoint "intended as a foundation checkpoint for further robot fine-tuning".

Works across robotsyesA

Developer claim: "post-trained, multi-embodiment MolmoAct2 model". Limit: "the model only works out of the box on the robot setups it was heavily trained on: SO-100, bimanual YAM, and Franka".

Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.

Notes

  • Ai2 calls MolmoAct 2 an open vision-language-action model for robot control; Ai2 also calls the family an action reasoning model.
  • Ai2 released MolmoAct 2 on 5 May 2026; MolmoAct 2-Think adds depth tokens for tasks that need 3D reasoning.

Sources

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.