What is it?
Some AI models can be downloaded and run on your own machines — open-weights models. Others are only available through a company's service. Open weights give you control and privacy, but the licence terms differ and aren't always as free as the word "open" suggests.
Also known as: open weights, open weight models, open source LLM, open-source AI, Llama licence, open vs closed models, self-hosted models, open source AI definition.
Technical definition
Open-weights releases publish parameters (and sometimes code), but training data and recipes may be undisclosed and licences may add use restrictions (field-of-use, size thresholds, attribution). The Open Source Initiative published an Open Source AI Definition (v1.0, 2024) with requirements beyond weights availability; whether a given release qualifies must be checked against current texts. Trade-offs vs closed APIs: control, privacy, customisation and cost structure versus capability, convenience and maintenance burden.
Entity type: concept · Freshness class: volatile.
When should I use it?
Choose open weights when you need control, privacy or customisation; closed APIs when you need the strongest capability with minimal operations.
Implementation patterns
- Read the licence and acceptable-use policy before adoption.
- Pin model revisions by hash.
- Keep a fallback hosted model.
- Scan downloaded artefacts and prefer safe formats.
Technologies and frameworks
Commonly associated: Hugging Face Hub, Llama family, Mistral family, Qwen family, OSI Open Source AI Definition. Names are examples of what exists, not recommendations, and projects change quickly.
Practical guide
- List requirements (privacy, cost, customisation).
- Shortlist models; read licences.
- Verify provenance and checksums.
- Evaluate on your tasks.
- Plan serving (Model Serving and Inference Infrastructure).
Common failure modes
- Assuming commercial use is allowed.
- Downloading unverified weights or pickled files from unknown sources.
- Ignoring that data provenance is undisclosed.
- Underestimating operating cost.
Troubleshooting
- Unsure about licence — Have legal review; record the licence version at download.
- Supply-chain worry — Use safetensors, verify checksums, download from official organisations.
Contrasted with
- Local AI vs Cloud AI: Privacy, Cost, Capability — A deployment decision that often follows the open/closed choice.
Questions this page answers
- What does open weights mean?
- Is an open-weights model open source?
- Open source vs closed source LLMs: which should I use?
- Can I use open-weights models commercially?
Typical goals: choose between open and closed models; understand licences.
Sources and verification
Status: not independently verified. This entry was authored on 2026-10-06 from general technical knowledge, without live checking of the sources below. Treat every source as a lead to confirm (identifier, title, authorship, current version) before publication or citation.
Freshness class: volatile — Products, versions, protocols, benchmarks, leaderboards, regulations and dates. Verify against the primary source before relying on specifics; re-review every 3 months.
Time-sensitive claims to verify before relying on them:
- Licence terms of specific models.
- Current OSI definition version and which releases qualify.
- Which models are currently released as open weights.
Canonical sources (unverified leads):
- Open Source Initiative, The Open Source AI Definition — opensource.org/ai
- Touvron et al., Llama 2: Open Foundation and Fine-Tuned Chat Models (2023) — arXiv:2307.09288
- Jiang et al., Mistral 7B (2023) — arXiv:2310.06825