TechNvidia

Explainer

Updated . Checked .

Which AI companies release open-weight models?

Key facts

The licenses range from Apache 2.0 to a publisher's own terms with conditions, and this page quotes each one from the publisher's own page.

As of Nov. 20, 2026, OpenAI, Meta, Google, Nvidia, Mistral AI, Qwen and DeepSeek have each published model weights, under licenses that range from Apache 2.0 to the publisher's own terms, and each license below is quoted from the publisher's own page, starting with OpenAI's gpt-oss-120b card. The table covers one model per publisher, plus a second Meta model, checked on that date. It ranks nothing and does not survey every model a publisher has released.

The license on each card

Publisher and modelLicense as the page states itOther details on the page
OpenAI, gpt-oss-120b"License: apache-2.0"; "Permissive Apache 2.0 license"117B total and 5.1B active parameters
Meta, Llama 4Llama 4 Community License Agreement, effective April 5, 2025Conditions on large licensees, naming and an acceptable use policy
Meta, Muse-Glimmer-30B"All artifacts are released under Apache 2.0"About 29.6B parameters including a vision encoder
Google, Gemma 3 27B IT"License: gemma"A user must accept conditions to access the files; terms are linked from the card
Nvidia, Nemotron 3 Nano 30B A3B"NVIDIA Nemotron Open Model License""This model is ready for commercial use"; 30B total and 3.5B active parameters
Mistral AI, Mistral Small 3.1 24B Instruct 2503"Apache 2.0 License: Open license allowing usage and modification for both commercial and non-commercial purposes."24B parameters
Qwen, Qwen3.5-27B"License: apache-2.0"27B parameters in the language model
DeepSeek, DeepSeek-V3Code under the MIT License; models "subject to the Model License"671B total and 37B activated parameters per token

The Qwen organization page says Qwen "refers to the large language model family built by Alibaba Cloud." The DeepSeek card says the "DeepSeek-V3 series (including Base and Chat) supports commercial use" and links to the Model License without reproducing its terms, so this page does not state them.

The same licenses grouped by type

License type on the cardPublishers and models in the table
Apache 2.0OpenAI gpt-oss-120b, Meta Muse-Glimmer-30B, Mistral Small 3.1, Qwen3.5-27B
The publisher's own licenseMeta Llama 4, Google Gemma, Nvidia Nemotron Open Model License, DeepSeek Model License (models) alongside MIT (code)

The cases a reader meets next

Meta's Llama 4 agreement says that if, on the release date, the monthly active users of a licensee's products or services were "greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta," which Meta "may grant to you in its sole discretion." It also says a distributor must "prominently display" "Built with Llama," and that a model improved with Llama materials and distributed must begin its name with "Llama," according to Meta's license page.

Google's card says the repository "is publicly accessible, but you have to accept the conditions to access its files and content," per the Gemma 3 27B IT card. The other model cards in the table show no such notice.

What changed and when

Meta's Llama 4 license took effect April 5, 2025, per Meta's license page. The Hugging Face blog posted on Aug. 10, 2026 describes Meta's Muse Glimmer as "released under the Apache 2.0 license," and Meta's card lists the authors as "Meta Superintelligence Lab." Meta's two models in the table therefore carry two different licenses. The gpt-oss paper listed on OpenAI's card is dated Aug. 8, 2025.

The page does not list publishers that release no weights, because no company page opened in this check states that of any company.

For what the term means and how it differs from the Open Source Initiative's definition, see What is an open-weight AI model?. Federal rules on chips are covered in Commerce Department adds export license rules for advanced chips in overseas data centers. Factsdesk reports what each page says and does not predict what any publisher or government will do.

Have another question?