Licensing · Commercial use
Local LLMs you can use commercially
"Open" does not always mean "free to ship". Some local models are fully permissive, some carry a user cap or terms to accept, and a few are research-only. Here they are, grouped by what their license actually allows.
- Free to ship
- 48
- With conditions
- 15
- Non-commercial
- 4
Free to ship 48
Apache-2.0, MIT and similar. Commercial use, modification and redistribution, no user cap.
- DeepSeek R1 671BMIT
- DeepSeek V3 671BMIT
- Qwen3 235B A22B 235BApache-2.0
- gpt-oss 120B 117BApache-2.0
- S Sarvam-105B 105BApache-2.0
- Mixtral 8x7B 46.7BApache-2.0
- Yi Yi 1.5 34B 34BApache-2.0
- Qwen2.5 32B 32BApache-2.0
- Qwen3 32B 32BApache-2.0
- DeepSeek-R1-Distill-Qwen 32B 32BMIT
- Qwen2.5 Coder 32B 32BApache-2.0
- Granite 4.0 H Small 32BApache-2.0
- Qwen3 30B-A3B 30.5BApache-2.0
- S Sarvam-30B 30BApache-2.0
- Mistral Small 3 24B 24BApache-2.0
- S Sarvam-M 24B 24BApache-2.0
- gpt-oss 20B 21BApache-2.0
- DeepSeek-V2-Lite 16BDeepSeek License · DeepSeek Model License permits commercial use.
- Phi-4 14B 14BMIT
- Qwen2.5 14B 14BApache-2.0
- Qwen3 14B 14BApache-2.0
- DeepSeek-R1-Distill-Qwen 14B 14BMIT
- Qwen2.5 Coder 14B 14BApache-2.0
- Mistral Nemo 12B 12.2BApache-2.0
- FN Falcon3 10B 10BFalcon LLM License 2.0
- Qwen2.5-VL 7B 8.29BApache-2.0
- Qwen3 8B 8BApache-2.0
- DeepSeek-R1-Distill-Llama 8B 8BMIT
- Mistral 7B 7BApache-2.0
- Qwen2.5 7B 7BApache-2.0
- DeepSeek-R1-Distill-Qwen 7B 7BMIT
- Qwen2.5 Coder 7B 7BApache-2.0
- Qwen3 4B 4BApache-2.0
- Phi-3.5-mini 3.8B 3.82BMIT
- Phi-4-mini 3.8B 3.8BMIT
- Qwen2.5-VL 3B 3.75BApache-2.0
- SmolLM3 3B 3BApache-2.0
- Granite 3.1 2B 2.53BApache-2.0
- SmolLM2 1.7B 1.7BApache-2.0
- Qwen3 1.7B 1.7BApache-2.0
- Qwen2.5 1.5B 1.54BApache-2.0
- Qwen2.5 Coder 1.5B 1.54BApache-2.0
- TL TinyLlama 1.1B 1.1BApache-2.0
- Qwen3 0.6B 0.6BApache-2.0
- Qwen2.5 0.5B 0.494BApache-2.0
- Qwen2.5 Coder 0.5B 0.494BApache-2.0
- SmolLM2 360M 0.362BApache-2.0
- SmolLM2 135M 0.135BApache-2.0
Free with conditions 15
Usable commercially, but with a caveat: a user cap, naming rules or a terms-of-use to accept.
- Llama 4 Scout 109BLlama 4 Community · Llama 4 Community License: free under 700M MAU, with use restrictions.
- Qwen2.5 72B 72BQwen License · Qwen License: free for commercial use under 100M monthly active users.
- Llama 3.3 70B 70BLlama 3.3 Community · Llama 3.3 Community License: free under 700M MAU.
- Gemma 2 27B 27BGemma · Gemma Terms of Use apply to commercial use.
- Gemma 3 27B 27BGemma · Gemma Terms of Use apply to commercial use.
- Gemma 3 12B 12BGemma · Gemma Terms of Use apply to commercial use.
- Llama 3.2 Vision 11B 10.7BLlama 3.2 Community · Llama 3.2 Community License: free under 700M MAU.
- Gemma 2 9B 9BGemma · Gemma Terms of Use apply to commercial use.
- GL GLM-4 9B 9BGLM-4 License · GLM-4 Model License: review terms for commercial use.
- Llama 3.1 8B 8BLlama 3.1 Community · Llama 3.1 Community License: free under 700M MAU.
- Gemma 3 4B 4BGemma · Gemma Terms of Use apply to commercial use.
- Llama 3.2 3B 3BLlama 3.2 Community · Llama 3.2 Community License: free under 700M MAU.
- Gemma 2 2B 2.61BGemma · Gemma Terms of Use apply to commercial use.
- Llama 3.2 1B 1BLlama 3.2 Community · Llama 3.2 Community License: free under 700M MAU.
- Gemma 3 1B 1BGemma · Gemma Terms of Use apply to commercial use.
Non-commercial only 4
Research or personal use only under their licenses. Not for a product you sell.
- CR Command R 35B 35BCC-BY-NC-4.0 · CC BY-NC 4.0: non-commercial use only.
- Qwen2.5 3B 3.09BQwen Research License · Qwen Research License: non-commercial use only.
- Qwen2.5 Coder 3B 3.09BQwen Research License · Qwen Research License: non-commercial use only.
- S Sarvam-1 2B 2BSarvam non-commercial · Non-commercial use only per the Sarvam license. Check the current HuggingFace model card for updates.
This is a plain-English summary to help you shortlist, not legal advice. Licenses change and have edge cases. Always confirm the current license on the model's HuggingFace page before you build on it.
FAQ
Can I use Llama models commercially?
Yes, with a condition. Llama models use Meta's community license, which is free for commercial use unless your product has more than 700 million monthly active users, at which point you must request a license from Meta. For almost everyone that cap never bites. Llama 4 adds further use restrictions on top of the cap, so check the Llama 4 Community License before shipping. This is why Llama sits in the conditional tier, not the fully permissive one.
What is the most permissive license for a local LLM?
Apache-2.0 and MIT are the cleanest: free for commercial use, modification and redistribution with no user cap. Most Qwen3 and Mistral models ship under Apache-2.0, and the DeepSeek-R1 distills ship under MIT, which is why they fill the permissive tier here.
Which local models can I not use commercially?
Command R 35B, Qwen2.5 3B, Qwen2.5 Coder 3B, Sarvam-1 2B are non-commercial only. Their licenses (the Qwen Research License, the Sarvam non-commercial license, and CC BY-NC) restrict them to evaluation and personal projects, not a product you sell.
Is this legal advice?
No. This is a plain-English summary of each model's stated license to help you shortlist. Licenses change and terms have edge cases, so confirm the current license on the model's HuggingFace page before you ship.
Sources
Licenses are summarized from each model's card, validated 2026-06-15. See methodology.