By Local AI Scout · Local · Models · Guide · Docs-only · Checked November 17, 2026
Short answer: Often yes, but read the license for the exact model you download. On the model cards we opened, Gemma 4, Qwen3.6, gpt-oss and Mistral Small 3.1 are tagged Apache 2.0, a permissive license that allows commercial use. Meta's Llama licenses allow commercial use with conditions: if you distribute Llama or build a product that contains it, you must include the license and display "Built with Llama," and companies above 700 million monthly active users need Meta's permission. Some open-weight models are non-commercial: Cohere's Command R7B is tagged CC BY-NC 4.0, and FLUX.1 [dev] carries a non-commercial license. The app you run them in has its own, separate license. This is general information, not legal advice.
This guide follows one reader: a small marketing agency that wants to run a local model for client drafts and possibly ship a chatbot inside a client's website.
Open weights is not the same as "free for business"
"Open weights" only means you can download the model files. What you may do with them is set by the license. Three questions decide most cases:
- Is commercial use allowed? Some licenses say no outright.
- Are you only using it, or distributing it? Many conditions start when you ship the model, or a product containing it, to someone else.
- Are there use restrictions? Some licenses attach an acceptable-use or prohibited-use policy that you must follow and pass on.
What the licenses we opened say
| Model (as listed on Hugging Face) | License tag | What that means in short |
|---|---|---|
| google/gemma-4-E4B-it, google/gemma-4-31B-it | apache-2.0 | Permissive; commercial use allowed |
| Qwen/Qwen3.6-27B, Qwen/Qwen3-8B | apache-2.0 | Permissive; commercial use allowed |
| openai/gpt-oss-20b | apache-2.0 | Permissive; commercial use allowed |
| mistralai/Mistral-Small-3.1-24B-Instruct-2503 | apache-2.0 | The card says "usage and modification for both commercial and non-commercial purposes" |
| Llama 3.1 and Llama 4 | Llama Community License | Commercial use with conditions (see below) |
| meta-llama/Llama-3.2-3B-Instruct | llama3.2 | Meta's own Llama 3.2 license; we didn't open the full text today ‡ |
| CohereLabs/c4ai-command-r7b-12-2024 | cc-by-nc-4.0 | Non-commercial |
| black-forest-labs/FLUX.1-dev | flux-1-dev-non-commercial-license | Non-commercial (image model) |
| black-forest-labs/FLUX.1-schnell | apache-2.0 | Permissive; note it differs from FLUX.1 [dev] |
Sizes and names within a family can carry different licenses, as the FLUX pair shows. Check the card for the exact file.
The Llama conditions in plain English
From the Llama 3.1 Community License (Llama 4's agreement, effective April 5, 2025, has the same clauses):
- If you distribute Llama or a product that contains it, include a copy of the agreement and prominently display "Built with Llama."
- If you use Llama to create or improve an AI model you distribute, its name must start with "Llama."
- Keep the attribution notice in copies you distribute, and follow Meta's Acceptable Use Policy.
- If your products had more than 700 million monthly active users on the release date, you need a separate license from Meta.
For a small business running Llama privately on its own computer, the distribution clauses don't come up. They matter when you ship it.
What about older Gemma models?
Our Gemma 4 guide notes that Gemma 1, 2 and 3 remain under Google's Gemma Terms of Use. Those terms (last modified April 1, 2026) require anyone who distributes Gemma or a derivative to pass on the use restrictions from Google's Prohibited Use Policy and give notice that they apply.
How to check any model in two minutes
- Open the model's page on Hugging Face. Hugging Face says repo owners specify the license in the model card's metadata; it shows as a tag near the top.
- If the tag says "other," open the linked license file and read it.
- Look for "non-commercial," "research only," user thresholds, attribution requirements and use policies.
- Check the app's license too. Our free local AI apps compared page covers Ollama, LM Studio, Jan and others.
- Save a copy of the license with the date you downloaded the model.
Choose a permissive model if…
- You'll put the model in something customers use, or you don't want to track conditions. Apache 2.0 models are the simplest. See best Ollama models and run Qwen locally.
Get advice first if…
- You plan to sell, host or embed a model for clients, fine-tune and redistribute it, or use a model with a custom or non-commercial license. We don't give legal advice; a qualified lawyer should read the actual terms.
Frequently asked questions
Is open weights the same as open source?
Not always. Open weights means you can download the model files. What you may do with them depends on the license, which ranges from permissive Apache 2.0 to custom terms and non-commercial licenses.
Can I use Llama in my business?
Meta's Llama licenses allow commercial use with conditions. If you distribute Llama or a product containing it, include the license and display "Built with Llama"; companies above 700 million monthly active users need Meta's permission.
Is Gemma 4 free for commercial use?
The Gemma 4 model cards we opened on Hugging Face are tagged Apache 2.0, which allows commercial use. Older Gemma models remain under Google's Gemma Terms of Use.
Do I need a lawyer to use a local model?
For private use under a permissive license, usually not. If you plan to ship, host or resell a model, or the license is custom or non-commercial, ask a qualified lawyer. This page isn't legal advice.
Sources
- Hugging Face model API license fields for the models listed in the table, opened November 17, 2026.
- Mistral Small 3.1 model card (https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503), opened November 17, 2026.
- Llama 3.1 Community License (https://www.llama.com/llama3_1/license/) and Llama 4 Community License (https://www.llama.com/llama4/license/), opened November 17, 2026.
- Gemma Terms of Use (https://ai.google.dev/gemma/terms), last modified April 1, 2026, opened November 17, 2026.
- Hugging Face Hub docs, "Licenses" (https://huggingface.co/docs/hub/repositories-licenses), opened November 17, 2026.
How we checked: a documentation-based guide, not a hands-on test. Sources opened November 17, 2026 unless stated. ‡ marks something we could not confirm on an official page. See our editorial policy.
