“Open model” can mean very different legal terms. The licence, not the download button, decides what you can build.
How it works
Open-weight models usually ship under one of two kinds of licence.
Standard software licences. Apache 2.0 and MIT are two examples. Apache 2.0 grants a “perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license”, plus a patent licence that ends for anyone who sues claiming the work infringes a patent. To redistribute, you must pass on the licence, mark changed files and keep the notices.
Custom model licences. A model maker writes its own terms. These can look permissive but add conditions on certain users or uses.
The Open Source Initiative’s Open Source AI Definition sets a stricter bar than “weights are public”. It asks for the freedom to “Use the system for any purpose and without having to ask for permission”, to study, modify and share it, and for data information, code and parameters under OSI-approved terms.
Here is what the cards of the models on BroadNet say, as checked on 15 November 2026:
| Model | Licence on the model card |
|---|---|
| DeepSeek V4.1 Flash | MIT: “This repository and the model weights are licensed under the MIT License.” |
| GLM 5.3 | Custom “GLM-5.3 License” from Z.AI |
| Qwen3.8 27B | apache-2.0 (card metadata) |
| Gemma 4 VL 26B | apache-2.0 (Google’s card) |
| Cohere Transcribe Arabic | Apache 2.0 |
| Cohere Transcribe | Apache-2.0, per BroadNet’s model page |
The GLM-5.3 License grants broad permission free of charge but defines “Model as a Service”. A licensee running such a business with combined revenue above USD 10 billion over 12 months must pass Z.AI’s security review before commercial use.
Why it matters
- Self-hosting and fine-tuning. If you plan to run, fine-tune or distill a model, the licence decides whether you may and what you must pass on.
- Outputs. Read what the licence says about outputs before you train other models on them.
- Vendor choice. An open-weight model under a permissive licence can be moved to another host or your own GPUs, which lowers lock-in. Our post on open-weight vs closed models covers that trade-off.
This page is a summary, not legal advice. Read the full licence text.
On BroadNet Inference
BroadNet’s Terms of Service (§7.3) say: “Models are open-weight models governed by their licences, named in our catalogue, or models trained by BroadNet.” They add that outputs are yours to use subject to those licences and the law, and that “DeepSeek V4.1 Flash remains open-weight when partner-hosted; its licence still applies.” Use through the hosted service also follows the partner restrictions in BroadNet’s Acceptable Use Policy.