EMBODIED REASONING SANDBOX

Molmo2-ER Image Inspector

Upload an image, ask a spatial or pointing question, and inspect the model's response and grounded coordinates.

Input image

Prompt

Max new tokens

32 512

Temperature (0 = deterministic)

0 1.5

Model response

Point-token visualization

Decoded point coordinates

Prompt ideas

Model: allenai/Molmo2-ER. Point prompts return <points coords=...> tokens, visualized above. Molmo2-ER is a perception backbone and does not produce robot-control action tokens.