Model licence The terms that decide what you may do with a downloaded model.
Short answer. A model licence is the legal permission attached to a set of model weights. It decides whether you may use the model commercially, modify it, share it and build on it. Apache 2.0 grants a perpetual, worldwide, royalty-free copyright licence and asks mainly for a copy of the licence and retained notices. The Llama 3.1 and 3.3 community licences add conditions: a licence request to Meta if your products had more than 700 million monthly active users, a “Built with Llama” notice, and an acceptable use policy. The Gemma terms of use point to a prohibited use policy, while Gemma 4 uses Apache 2.0. This is general information, not legal advice; read the licence for the exact version you use.
01
Licences compared, as of October 2026
Licence
Models that use it (as stated on their pages)
Main conditions
Apache 2.0
Qwen3, gpt-oss, Gemma 4; also the Qwen-based DeepSeek distilled models alongside MIT
Copyright and patent licence with no charge. Give recipients a copy, keep notices, mark changes. The patent licence ends if you sue over patents in the work. No trademark rights. Provided as is
MIT
DeepSeek-R1 distilled models, per their model cards
A short permissive licence; see the text in the repository
Llama 3.1 Community License and Llama 3.3 Community License
Llama 3.1, Llama 3.3
Show “Built with Llama” and a notice file when you distribute or offer a product that contains the materials. Follow the acceptable use policy. Ask Meta for a licence if monthly active users were over 700 million on the release date. A distributed model improved with Llama outputs must start its name with Llama
Gemma Terms of Use
Gemma 1 to 3 and other variants listed in its appendix
Pass the use restrictions on to anyone you distribute to. Follow the Gemma Prohibited Use Policy. Google reserves the right to restrict use, remotely or otherwise, that it reasonably believes breaches the terms
02
What to check before you rely on one
The version: each model release has its own licence text, with its own date and policy link.
Commercial use and size thresholds: some licences set conditions based on how large your user base is.
Notice and attribution duties: what you must display or include when you pass the model on.
The acceptable use policy: banned uses that the licence incorporates by reference.
Derived models: fine-tunes, adapters and distilled copies usually carry the base model’s terms.
Remote restriction clauses: the Gemma terms reserve a right to restrict use remotely or otherwise. Ask what that means for a system that must work offline.
Keep a copy of the licence file with the weights, with the date you downloaded them.
03
Internal use and passing it on
Duties differ between using a model inside the office and giving others access to it. The Llama licences apply their notice and naming conditions when you distribute or make the materials available, or a product or service containing them. The Gemma terms define distribution broadly, to include making Gemma or its functionality available as a hosted service through an API or web access. A firm that gives clients access to an assistant built on a model should read this part of the licence with a lawyer.
A licence on weights is separate from rights in the data a model was trained on and in what it writes. The Open Source Initiative’s definition of open source AI asks for sufficiently detailed training data information as well as the weights and code, which is why many open-weight releases are not open source AI.
04
Questions to ask a vendor
Which licences apply to each model you ship, and for which versions?
Do those licences allow my office’s use, including serving many named users?
Who is responsible for notices and acceptable-use compliance, you or me?
What happens if a model is withdrawn or its licence changes?
Can I keep the weights and their licence files if I leave?
05
Where ANVE stands
ANVE’s security page says every order includes a perpetual licence for the software delivered with the unit, that there is no remote kill switch, and that the box keeps working if ANVEAI shuts down. That covers ANVE’s own software. Each open model on the box keeps the licence its maker gave it, so the questions above still apply to the models.
06
Questions people ask
Can I use Llama commercially?
The Llama 3.1 and 3.3 licences allow commercial use with conditions, including a licence request to Meta above 700 million monthly active users, a notice, and an acceptable use policy. Read the text for your case. This is general information, not legal advice.
Is Apache 2.0 enough for commercial use?
Apache 2.0 grants a perpetual, worldwide, no-charge, royalty-free copyright licence to reproduce and distribute the work, with conditions on notices. It is a permissive licence. This is general information, not legal advice.
Does a new Gemma model use the same terms?
Not necessarily. Google’s terms page says Gemma 4 has its own licence, and its licence page shows Apache 2.0. Earlier Gemma models are listed under the Gemma Terms of Use.
Do fine-tuned models need a new licence?
A fine-tune generally remains subject to the base model’s licence. Llama’s also requires the word Llama at the start of the name of a distributed model improved with its outputs.
07
What is not real yet
No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.
