← AI Terminology
Open Weights vs Open Source
Open weights means model parameters are downloadable for use; open source further implies usable source, training code, and licenses that meet open-source definitions — not always the same thing.
The distinction matters legally and practically in the LLM ecosystem.
The distinction matters legally and practically in the LLM ecosystem.
Why It Matters in AI
Many “open” models release weights under research or restricted licenses without full training stacks. Practitioners must read licenses (commercial use, copyleft) and not assume OSI open source. Strategy and compliance depend on the difference.
Key Points
| Aspect | Description |
|---|---|
| Ops | License review before enterprise deploy |
| Risk | Calling restricted weights “open source” incorrectly |
| Related | Frontier models, Hugging Face hubs |
| Examples | Some weights-only releases vs fully open stacks |
| Open source | Code + freedoms per OSI-style definitions |
| Open weights | Checkpoints available; license may restrict use |
Simple Analogy
Getting a free cake versus getting the recipe, kitchen rights, and permission to sell your own bakery line — both “open,” very different freedoms.
Common Usage Examples
- Read MODEL LICENSE before commercial use
- Compare Llama license vs Apache-2.0 models
- Prefer fully open for air-gapped forks
- Legal review of weight licenses
Summary
In short: Open weights give you parameters; open source gives broader freedoms — do not conflate them when choosing models.