Open-weight or open-source AI? What the labels mean

Many “open” models let you download the weights but not much else. The difference matters when you choose what to build on.

Team Inspiropedia

2 min read

Key points

  • Open-weight means the trained parameters can be downloaded and run yourself.
  • Open source, as the Open Source Initiative defines it for AI, requires more than the weights.
  • Always read the licence: many popular models carry custom terms.

“Open” is one of the most used and least precise words in AI. A model released as open-weight gives you the trained numbers, the weights, so you can run it on your own hardware, fine-tune it and keep your data in-house. That is valuable. It is not the same as open source.

What open source means

The Open Source Initiative published its Open Source AI Definition 1.0 in October 2024. In short, it says a system should let you use it for any purpose, study how it works, modify it and share it, and that this requires the weights, the code used to train and run it, and sufficiently detailed information about the training data. Many open-weight releases do not meet that bar.

Licences matter more than labels

Some open-weight models use standard licences such as Apache 2.0 or MIT. Others use custom community licences that may limit commercial use above a certain scale, restrict particular uses or require attribution. Before building a product, read the licence and check it fits your use.

Choosing between open and hosted models

  • Control and privacy. Running a model yourself keeps data inside your own systems.
  • Cost. At high, steady volume self-hosting can be cheaper; at low volume a hosted API often is.
  • Effort. You take on hosting, monitoring, security updates and safety testing.
  • Performance. The best hosted models may still lead on some tasks. Test on your own data rather than relying on leaderboards.