Open-weight AI models and the licenses that come with them
Eight of the 9 AI companies we checked publish open-weight models, and 7 of the 11 latest releases we list come under Apache 2.0 or MIT, which allow commercial use with few conditions. The exceptions are Llama 4, Mistral Medium 3.5, Qwen3.8 Flash-Next and Grok 2, whose licenses add conditions by company size, business type or a usage policy. Anthropic's pages describe no open-weight Claude model.
Every "uncensored" model you can download started as someone's open weights, and those weights come with a license. Some licenses let you do almost anything; others add a usage policy, a size limit or a ban on certain businesses. We read the official model pages and license texts of the latest open-weight releases from 9 AI companies and put the terms side by side.
Open-weight models and their licenses
From each vendor's official model page and the license text itself, checked 2026-10-02. Click a name to open the source.
| Meta · Muse Glimmer 30B | Apache 2.0 | Allowed | None | 2026-08 | |
|---|---|---|---|---|---|
| Meta · Llama 4 | Llama 4 Community License | Allowed; above 700 million monthly users, ask Meta for a license | Yes, attached | 2025-04 | “is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta” |
| Mistral AI · Mistral Medium 3.5 | Modified MIT License | Allowed, except for companies with more than $20 million monthly revenue | None | 2026-03 | “Open-source license for both commercial and non-commercial use with exceptions for companies with large revenue.” |
| Mistral AI · Mistral Small 4 | Apache 2.0 | Allowed | None | 2026-03 | |
| DeepSeek · DeepSeek V4.1 Flash | MIT | Allowed | None | 2026-09 | |
| Google · Gemma 4 | Apache 2.0 | Allowed | None | 2026-03 | |
| OpenAI · gpt-oss-120b, gpt-oss-20b | Apache 2.0 | Allowed | Only: comply with the law | 2025-08 | “By using OpenAI gpt-oss-120b, you agree to comply with all applicable law.” |
| Alibaba (Qwen) · Qwen3.8 27B | Apache 2.0 | Allowed | None | 2026-08 | |
| Alibaba (Qwen) · Qwen3.8 Flash-Next | Qwen Community License 1.0 | Allowed; large products must show the model name, and model-as-a-service businesses need a separate license | None | 2026-08 | “the licensee shall obtain a separate license from Qwen before Using the Software or its derivative works for any commercial purpose” |
| Microsoft · Phi-4 reasoning vision 15B | MIT | Allowed | None | 2026-03 | |
| xAI · Grok 2 | xAI Community License | Allowed if you follow xAI's usage policy; not for training other general-purpose models | Yes, attached | 2025-08 | “For commercial use solely if you and your affiliates abide by all of the guardrails provided in xAI's Acceptable Use Policy” |
| Anthropic | No open-weight model on its pages | — | — | — | |
| Uncensored AI Chat | Open-weight models | On our own GPU servers | Not sent to OpenAI, Google, Anthropic or any other AI company |
How we checked: we read each model's page in the vendor's own verified organization on Hugging Face or on the vendor's site, and the license text that comes with the weights, on the date above. The table lists each vendor's latest open-weight release and, where a vendor has two licenses in use, one model under each. A vendor may publish more models than listed. We update the table every month.
Read next
On this page
How to read the table
Each row is one model as named by its vendor, with the license that comes with its weights. The commercial-use column summarizes what the license says about using the model in a product, and the next column says whether the license attaches a usage policy. Where a license adds conditions, the last field quotes it word for word.
Apache 2.0 and MIT: the permissive end
Apache 2.0 and MIT let you use, change and sell products built on the model, as long as you keep the license and copyright notices. Meta's Muse Glimmer, Mistral Small 4, DeepSeek V4.1 Flash, Google's Gemma 4, OpenAI's gpt-oss, Qwen3.8 27B and Microsoft's Phi-4 all come this way. Gemma 4 is a change for Google: its license page shows Apache 2.0, where earlier Gemma models came with Google's own terms of use.
Community licenses: conditions by size and business
Llama 4's license asks companies with more than 700 million monthly active users to request a separate license from Meta. Mistral Medium 3.5 excludes companies with more than $20 million monthly revenue. Qwen3.8 Flash-Next asks large products to show the model name and requires a separate license for model-as-a-service and AI work assistant businesses.
Usage policies that travel with the weights
Llama 4 and Grok 2 tie their licenses to an acceptable use policy, so the vendor's rules on use come with the weights. xAI also forbids using Grok 2 or its outputs to train other general-purpose models and asks for a "Powered by xAI" notice. OpenAI's usage note for gpt-oss asks only that you comply with the law.
What this means for uncensored models
A fine-tuned or "uncensored" model is a modified copy of the original weights, so the conditions of the original license still apply to it. A model built on Apache 2.0 or MIT weights carries few conditions; one built on Llama 4 or Grok 2 still carries their usage policy. This chat runs open-weight models on its own servers, and your messages are not sent to other AI companies.
Questions people ask first
Is Llama 4 open source?
Its weights are open, but its license is Meta's own Llama 4 Community License, with a 700 million monthly user threshold and an acceptable use policy, rather than a standard open-source license.
Which open-weight models allow commercial use?
All the models in the table allow commercial use in some form. Those under Apache 2.0 or MIT allow it without extra conditions; Llama 4, Mistral Medium 3.5, Qwen3.8 Flash-Next and Grok 2 add conditions.
Does Anthropic release open-weight models?
Anthropic's pages we read describe no open-weight Claude model.
Does an uncensored fine-tune change the license?
No. A fine-tuned model is built from the original weights, so the original license's conditions, including any usage policy, still apply.
How often is this updated?
Every month. The date we last read the sources is shown above the table.
Open-weight models on our own servers, no account. Open the chat and ask.
Start uncensored chat