Open weights AI models differ from true open source

There is a critical distinction between AI models released with ‘open weights’ and those that are truly open source. While many companies market their models as open, they often only provide the weights—the numerical parameters learned during training that allow a model to function and be executed locally using tools like Ollama or LM Studio.

True open source AI, as defined by the Open Source Initiative (OSI) in its Open Source AI Definition 1.0, requires much more than just downloadable parameters. To meet this standard, a provider must supply the complete recipe: the full code used to train and run the system, and sufficient information regarding the training data, including its origin and filtering methods.

While open weights allow users to run models without sending data to external servers and permit some level of fine-tuning, they do not grant the four essential freedoms of free software: the ability to use the system for any purpose, study how it works, modify it, and share it freely. This distinction is vital for businesses intending to integrate AI into commercial products or modify the underlying technology.