Molmo 2
An open-weight multimodal vision-language model that supports image understanding and natural language interaction. Its weights can be downloaded, fine-tuned and redistributed, making it suitable for adaptation to vertical domains such as science.
Timeline (last 90 days)
MIT researcher Kai Van Brunt used the open-source Molmo 2 model to build Molmo2Fish, an experimental model that can track salmon from sonar imagery.
The open-weight release of Molmo 2 allows researchers to inspect the model's internal representations, fine-tune it for a specific domain, and embed the model into their own data pipeline.
Molmo 2 was mainly trained on internet-scale images such as everyday photos, charts and documents.
The open-sourcing of a general-purpose multimodal model spawned a cross-disciplinary scientific application prototype within a few months.
All Facts (4)
MIT researcher Kai Van Brunt used the open-source Molmo 2 model to build Molmo2Fish, an experimental model that can track salmon from sonar imagery.
50%UnverifiedThe open-weight release of Molmo 2 allows researchers to inspect the model's internal representations, fine-tune it for a specific domain, and embed the model into their own data pipeline.
50%UnverifiedMolmo 2 was mainly trained on internet-scale images such as everyday photos, charts and documents.
50%UnverifiedThe open-sourcing of a general-purpose multimodal model spawned a cross-disciplinary scientific application prototype within a few months.
50%