Open Weights — AI Glossary

Reference details

Related terms

• Closed Model
• Fine-Tuning
• Quantization
• GGUF

Example: Read the licence, not the download button

Two models both offer a weights download. One is under Apache 2.0: use it commercially, modify it, redistribute it, no threshold. The other is under a community licence that grants broad commercial use but requires a separate licence from the vendor once your product exceeds a stated size — Meta's Llama licence sets that at 700 million monthly active users — and adds acceptable-use terms that travel with the weights. Both are “open weights”. Only one is open source by the OSI definition, and neither publishes the training data.

Commonly confused with

Open weights is a distribution posture, not a licence and not a claim about openness of process. Open source additionally implies the licence permits any use without field restrictions; open data implies the corpus is published. Most models called “open source” in press coverage are open weights only.

When to use it

Reach for it when:

• Data residency or air-gapped deployment, where no request may leave your boundary
• Fine-tuning, quantizing or distilling on your own domain data
• Cost control at sustained high volume, where amortised hardware beats per-token pricing

Reach for something else when:

• Small or spiky volumes — self-hosting idle capacity costs more than an API
• Where you need the vendor's evaluation, safety tuning and incident response as part of the deal
• Before legal has read the licence: acceptable-use terms and thresholds bind downstream users too

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Open Weights — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path