An open model may release weights, code, architecture, data information, or documentation so others can inspect, run, modify, or redistribute some parts.
Openness is a spectrum. Check exactly what is available, the license, use restrictions, training transparency, dependencies, and practical compute needed.
Example
Weights available, data absent
A model may be downloadable while its training dataset and full recipe remain undisclosed.
Important limit
Open is not automatically safe, fair, cheap, or open source
Licenses and definitions differ; access can enable research and local control while also enabling misuse.
Interactive model
Separate before deciding
Selected layerOpen weights
Parameters can be downloaded.
Read the complete text alternative
Open weightsParameters can be downloaded.
Open sourceUsually concerns source code rights; applying the term to models is debated.
Useful distinctions
Separate the questions before judging
Open weights
Parameters can be downloaded.
Open source
Usually concerns source code rights; applying the term to models is debated.