# Open Model Licenses

> What the license on an open model lets you do. MIT and Apache 2.0, custom licenses with limits, and fully open models like Olmo.

**Course:** [AI Models and Providers](https://learn.benibix.com/learn/ai-models-providers)  
**Time:** 12 min

**After this lesson you can:**

- Tell open weights from fully open
- Say which licenses allow paid use
- Check a license before you ship

## What an open model's license lets you do

Every open-weight model comes with a license, and the license decides what you may do with it. MIT and Apache 2.0 let almost anyone use, change and sell with the model. Others add limits. The Llama 2 license asked any firm with over 700 million monthly users to get Meta's permission first.

Open comes in three levels. A closed model shares nothing you can download. An open-weight model shares the trained file. A fully open model shares the file, the data it learned from, and the code and steps used to train it. Most well-known open models stop at the second level.

The Allen Institute for AI, or Ai2, is a nonprofit lab in Seattle. A co-founder of Microsoft set it up in 2014. Its Olmo models are fully open. In February 2024 it shared the first Olmo with its code, its weights and its whole training set. Researchers use it to see how each step of training changes a model.

NVIDIA's Nemotron family is open in much the same way. Its maker shares the weights, much of the data and the recipes it used. It comes in three sizes. Nemotron Nano is for small, fast jobs, Nemotron Super for harder work and Nemotron Ultra for the hardest. It is tuned for its maker's chips, but the files run on others too.

Permissive licenses ask very little. MIT and Apache 2.0 let you use a model in a paid product, as long as you keep the license notice with it. DeepSeek-R1, the GLM models and Phi-4 use MIT. gpt-oss, Gemma 4 and most Qwen models use Apache 2.0.

Custom licenses add rules. Some limit who may use the model, as the Llama 2 user cap did. Some ban certain uses outright. Some split one family in parts. One FLUX picture model is under Apache 2.0, another allows only non-commercial use, and the strongest is never shared at all.

People treat a free download as free to use however they like. The file is free, but the license still binds you. It can change between versions of one family, too. Check the license on the exact version you ship, and keep a copy of it with your project.

## Common licenses on open models

| License | Paid use | Examples |
| --- | --- | --- |
| MIT | Yes | DeepSeek-R1, Phi-4, GLM |
| Apache 2.0 | Yes | gpt-oss, Gemma 4, most Qwen models |
| Llama 2 license | Yes, under a user cap | Llama 2 |
| Non-commercial | No | Some FLUX models |

![Two boxes side by side. Open weights: only the model file is shared. Fully open: the model file, the training data and the training code and recipes are all shared. Olmo and Nemotron sit in the fully open box.](https://learn.benibix.com/diagrams/ai9-fully-open.svg)

### Example: Jack checks before he ships

Jack's startup will sell a photo app for $4 a month. The picture model he likes best allows only non-commercial use.

He finds a second model under Apache 2.0. It scores 9 of 10 on his test photos against 10 of 10, and he ships with it.

**Words in this lesson:** [Software License](https://learn.benibix.com/term/development/software-license), [Permissive License](https://learn.benibix.com/term/development/permissive-license), [MIT License](https://learn.benibix.com/term/development/mit-license), [Apache License 2.0](https://learn.benibix.com/term/development/apache-license), [Allen Institute for AI (Ai2)](https://learn.benibix.com/term/ai/ai2), [Olmo](https://learn.benibix.com/term/ai/olmo), [Nemotron](https://learn.benibix.com/term/ai/nemotron)

**Question:** Does an open-weight model always let you use it in a paid product?

**Answer:** No, it depends on the license of that model. MIT and Apache 2.0 allow paid use. Other licenses add caps, ban uses or forbid business use, so you read the license on the exact model.

---

Source: https://learn.benibix.com/learn/ai-models-providers/open-model-licenses
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
