Open LLM Licence Checker
“Open weights” does not mean “open source”, and it certainly does not mean you can ship it. This checks every model in our directory against its licence family and spells out the obligations that usually surface far too late — the attribution you owe, the naming rules, and the models you simply cannot sell.
Not legal advice. This is a practical summary for engineers, not legal advice. Licence terms change and only the licence text itself is authoritative. Read the linked licence before shipping commercially, and take legal advice where the stakes warrant it.
| Model | Licence | Commercial use |
|---|---|---|
| Cohere Command R 35BCohere | CC-BY-NC-4.0 | Not commercial |
| Mistral Codestral 22BMistral AI | Mistral Non-Production | Not commercial |
| DeepSeek R1 Distill Llama 70BDeepSeek AI | Llama 3.3 Community | Conditions apply |
| DeepSeek R1 Distill Llama 8BDeepSeek AI | Llama 3.1 Community | Conditions apply |
| DeepSeek V3 671B MoEDeepSeek AI | DeepSeek Model License (code MIT) | Conditions apply |
| Google Gemma 2 27BGoogle DeepMind | Gemma Terms of Use | Conditions apply |
| Google Gemma 2 9BGoogle DeepMind | Gemma Terms of Use | Conditions apply |
| Llama 3.1 8B InstructMeta AI | Llama 3.1 Community | Conditions apply |
| Llama 3.3 70B InstructMeta AI | Llama 3.3 Community | Conditions apply |
| Llama 4 Scout (17B active, 16 experts)Meta | Llama 4 Community License | Conditions apply |
| Meta Llama 3.2 3B InstructMeta AI | Llama 3.2 Community | Conditions apply |
| Qwen 2.5 72B InstructAlibaba Cloud | Qwen License | Conditions apply |
| DeepSeek R1 Distill Qwen 14BDeepSeek AI | Apache 2.0 | Commercial OK |
| DeepSeek R1 Distill Qwen 32BDeepSeek AI | Apache 2.0 | Commercial OK |
| DeepSeek V4-FlashDeepSeek AI | MIT | Commercial OK |
| Gemma 4 12BGoogle DeepMind | Apache 2.0 | Commercial OK |
| Gemma 4 26B-A4B (MoE)Google DeepMind | Apache 2.0 | Commercial OK |
| Gemma 4 31BGoogle DeepMind | Apache 2.0 | Commercial OK |
| Microsoft Phi-4 14BMicrosoft | MIT | Commercial OK |
| Mistral NeMo 12B InstructMistral AI & NVIDIA | Apache 2.0 | Commercial OK |
| Mistral Small 3 24BMistral AI | Apache 2.0 | Commercial OK |
| OpenAI gpt-oss-120bOpenAI | Apache 2.0 | Commercial OK |
| OpenAI gpt-oss-20bOpenAI | Apache 2.0 | Commercial OK |
| Qwen 2.5 Coder 14B InstructAlibaba Cloud | Apache 2.0 | Commercial OK |
| Qwen 2.5 Coder 32B InstructAlibaba Cloud | Apache 2.0 | Commercial OK |
| Qwen 2.5 Coder 7B InstructAlibaba Cloud | Apache 2.0 | Commercial OK |
| Qwen3 8BAlibaba Qwen | Apache 2.0 | Commercial OK |
| Qwen3-Coder 30B-A3B InstructAlibaba Qwen | Apache 2.0 | Commercial OK |
| Qwen3.6 35B-A3B (MoE)Alibaba Qwen | Apache 2.0 | Commercial OK |
| Qwen3.8 27BAlibaba Qwen | Apache 2.0 | Commercial OK |
What each licence actually requires
Apache License 2.0
Commercial use permittedA standard permissive open-source licence with an explicit patent grant.
What you must do
- Include a copy of the licence with any distribution
- State any significant changes you made to the files
- Preserve copyright, patent, trademark and attribution notices, including any NOTICE file
MIT License
Commercial use permittedThe most permissive common licence. Almost no obligations beyond attribution.
What you must do
- Include the copyright notice and licence text in copies or substantial portions
Llama Community License
Commercial use permitted, with conditionsMeta calls it a community licence, not an open-source one. Commercial use is allowed below a scale threshold, but it carries naming and attribution duties most teams miss.
What you must do
- Prominently display "Built with Llama" on a related website, user interface, blogpost, about page or product documentation
- Include "Llama" at the beginning of the name of any AI model you train or fine-tune from Llama materials
- Include the licence and attribution notice with any redistribution
- Comply with Meta's Acceptable Use Policy
Easy to miss
- If your products exceed 700 million monthly active users, you must request a separate licence from Meta — it is not granted automatically
Gemma Terms of Use
Commercial use permitted, with use restrictionsGoogle permits commercial use, but the terms carry a Prohibited Use Policy that travels with the model and binds anyone you pass it to.
What you must do
- Pass the same terms and use restrictions to anyone you distribute the model or a derivative to
- Include a copy of the terms and note any modifications
- Comply with the Gemma Prohibited Use Policy
Easy to miss
- Google reserves the ability to restrict use it considers in breach of the Prohibited Use Policy, which is a broader hook than a standard open-source licence
Creative Commons BY-NC 4.0
Not for commercial useThe NC means non-commercial. Fine for research, prototypes and personal projects; not for anything that makes money without a separate licence from the rights holder.
What you must do
- Attribute the original creator
- Indicate if you made changes
Easy to miss
- "Non-commercial" covers more than direct sales — an internal tool at a for-profit company is usually commercial use
- Several strong embedding and reranker models ship under this licence, so RAG stacks trip over it often
Mistral Non-Production Licence
Not for production or commercial useResearch and evaluation only. Production use requires a commercial licence from Mistral.
What you must do
- Retain the licence and attribution notices
Easy to miss
- Internal production deployments count as production — contact Mistral for a commercial licence
- Not all Mistral models use this; many are Apache 2.0, so check per model rather than assuming
Qwen LICENSE AGREEMENT
Commercial use permitted, with conditionsAlibaba's own licence, not Apache. Commercial use is allowed, but above 100 million monthly active users you must request a separate licence. Note Qwen splits this by model size — several Qwen 2.5 models really are Apache 2.0, while the 72B is not.
What you must do
- Retain this notice in a "Notice" text file with any distribution: "Qwen is licensed under the Qwen LICENSE AGREEMENT, Copyright (c) Alibaba Cloud. All Rights Reserved."
- If you train, fine-tune or improve a distributed model using it, prominently display "Built with Qwen" or "Improved using Qwen" in the product documentation
Easy to miss
- Above 100 million monthly active users you must request a licence: "You cannot exercise your rights under this Agreement without our express authorization"
- No trademark licence is granted beyond the required notices
- Easy to assume the whole family is Apache 2.0 because most of it is — check the specific repo
DeepSeek Model License
Commercial use permitted, with use restrictionsPermissive on the IP grant, but it carries use-based restrictions that must travel with every derivative. The code repository is MIT; the model weights are not.
What you must do
- Include the use-based restrictions as an enforceable provision in any agreement governing the model or its derivatives, and give notice to downstream users
- Give any third-party recipient a copy of the licence
Easy to miss
- A derivative may be released under different terms but must include "at minimum - the same use-based restrictions"
- "Distribution" explicitly includes hosting as a service - "e.g. API-based or web access"
- The MIT badge on the repo covers LICENSE-CODE. The weights are governed by LICENSE-MODEL
Common questions
- Can I use Llama models commercially?
- Yes, below a scale threshold. The Llama Community License permits commercial use, but if your products exceed 700 million monthly active users you must request a separate licence from Meta. You must also prominently display "Built with Llama" and begin the name of any model you fine-tune from Llama with "Llama".
- Is Apache 2.0 safe for a commercial product?
- Yes. Apache 2.0 is a standard permissive open-source licence with an explicit patent grant. You keep the copyright and attribution notices, state significant changes, and ship the licence with any distribution. Most current open-weights models, including the Qwen3 family, use it.
- What does CC-BY-NC mean for a model?
- The NC stands for non-commercial, so you cannot use the model in anything that makes money without a separate licence from the rights holder. This catches teams out in RAG projects in particular, because several strong embedding and reranker models ship under it.
- Does an internal company tool count as commercial use?
- Under a non-commercial licence, usually yes. Non-commercial terms generally cover use by or for a commercial entity, not just direct sale. If the model is supporting a for-profit business, treat it as commercial use and get a licence.
- Are all Mistral models non-commercial?
- No, and this is worth checking per model. Many Mistral releases are Apache 2.0. Others use the Mistral Non-Production Licence, which restricts use to research and evaluation and requires a commercial licence for production.
Licence terms checked against the publishers’ own licence pages on 2026-10-03. Licences change — re-read the linked text before you ship.
Related guides
Where licensing bites in practice, and how to pick models around it.
- Best Local Embedding Models for RAGDimensions, licences and what vector storage really costs.Read the guide
- Best Local LLM Runners: Ollama vs LM Studio vs Jan vs llama.cpp vs vLLMWhich runner suits your hardware and workflow.Read the guide
- How Much RAM and VRAM Do You Need to Run Local LLMs?Model weights, KV cache and OS overhead, tier by tier.Read the guide