8 of 17 open-weights AI models ship under a license of their own, not MIT or Apache, including 6 of the top 10

8 of the 17 open-weights models on ArtificialWatch's list are released under a license of their own rather than MIT or Apache 2.0, including 6 of the 10 that rank highest. 6 use MIT, among them the top-ranked open model, DeepSeek V4.1 Flash. The 3 under Apache 2.0 rank lowest.

BY LICENSE

- A license of their own (8): Kimi K3.1 Pro, GLM-5.3, MiniMax M3.1 Flash, Step 5 Preview, Kimi K3, Mistral Large 4, Muse Spark Open and MiniMax M3.

- MIT (6): DeepSeek V4.1 Flash, MiMo-V2.6 Pro, GLM-5.2, GLM-5.3 Flash, DeepSeek V4 Pro and Ling 3.1 Flash.

- Apache 2.0 (3): Beam, Inkling and Inkling Small, which rank 43rd to 45th of the 45 models on the bench index.

AT THE TOP

DeepSeek V4.1 Flash, the highest-ranked open model at 11th overall, is MIT. The next, Kimi K3.1 Pro at 12th, uses Moonshot's own license, as does GLM-5.3 at 20th. Of the 10 highest-ranked open models, 4 are MIT and 6 carry their own terms.

ONE LAB'S SWITCH

Zhipu AI released GLM-5.2 under MIT and its successor GLM-5.3 under a license of its own; its smaller GLM-5.3 Flash stayed MIT.

WHAT THIS DOES NOT SAY

A license of a lab's own can be close to MIT or far from it; this count reads each repo's license tag, not the terms. Meta's Muse Spark Open also asks for manual approval before download. Read the license in each repo before building on a model.

HOW WE COUNTED

From ArtificialWatch's open-weights list (artificialwatch.com/open-weights), whose 17 Hugging Face repos were all re-read on November 5, 2026: 'a license of their own' means the repo's license tag is 'other' with the license's own name. Ranks are on ArtificialWatch's bench index of 45 models, read November 5; Ling 3.1 Flash is not yet ranked.