Free model licenses

This page lists the third-party weights offered as AI Models and the protocol that comes with each license. It is not legal advice. The vendor's license is what binds you. Omnific does not grant those rights.

This guide covers weights that are free to run and whose published license allows commercial use when you follow that license. Hosted model APIs are separate services. A consumer chat subscription is not an API key; each API has its own terms and billing.

Apache License 2.0

Commercial use is allowed. If you redistribute the weights or a derivative, give recipients a copy of the license, keep notices, mark files you changed, and do not use the vendor's trademarks as an endorsement. This covers Qwen2.5, Qwen2.5-Coder, Qwen3, Qwen3-Coder, Qwen3.5-9B, Mistral 7B, Moondream (confirm the tag), gpt-oss-20b (plus OpenAI's usage policy), and Gemma 4.

Apache License 2.0 · Gemma 4 · gpt-oss

MIT License

Phi-4 and its reasoning variants, and DeepSeek-R1. Keep the copyright and permission notices. A Qwen-based distill also stays subject to the Qwen license.

Gemma 2 and Gemma 3

Gemma Terms of Use, including the Prohibited Use Policy. Redistribution has to carry the agreement, the use restrictions, and the notice file. Terms · Prohibited uses

Llama 3.1 and Llama 3.2

Commercial use is allowed unless, in the month before that version's release, you or your affiliates already had more than 700 million monthly active users. Above that line you need a separate license from Meta. Follow the Acceptable Use Policy. If you distribute or make available Llama materials, derivatives, or a product or service containing them, follow the license-copy, notice, and “Built with Llama” requirements. These are conditional obligations for the relevant Llama-containing offering, not a blanket badge requirement for unrelated products. Check the exact model version and how it is distributed or made available. Llama 3.1 Community License · Llama 3.2 Community License

NVIDIA Nemotron open weights

Nemotron Mini, Nemotron Omni, and Nemotron 3 Nano Omni follow the NVIDIA Open Model License. A hosted NIM API is not one of these seats.