interconnects.ai web signal

Open-model licenses split as Chinese labs add revenue gates

TL;DR

  • Zhipu's GLM-5.3-Flash moves from MIT to a custom license that gates commercial use for Model-as-a-Service operators above $10 billion in affiliate revenue.
  • Motif Technologies shipped Motif-3 under MIT, following earlier Motif-2.6B and Motif-2-12.7B releases, with what Interconnects calls impressive scores for its size.
  • Google and Meta have moved their open releases to Apache 2.0 while Kimi K3 and MiniMax M3 require commercial agreements above revenue thresholds.

Zhipu has swapped the MIT license on its GLM-5.3-Flash release for a custom agreement that requires any Model-as-a-Service operator with more than $10 billion in aggregate affiliate revenue to "pass Z.AI's security review before using the Software or its derivative works for any commercial purpose," according to a roundup by Florian Brand and Nathan Lambert at Interconnects.

The threshold sits high, but the post flags that "affiliates" is undefined in the license text, which "creates barriers to adoption" for exactly the large operators the clause is written for.

GLM-5.3-Flash arrives in a batch alongside Motif Technologies' MIT-licensed Motif-3, which the authors credit with "impressive scores for its size" and describe as "innovation in their model training with very limited resources," building on earlier Motif-2.6B and Motif-2-12.7B releases. Tencent's Hy4-preview lands in the same roundup, called competent but with an "issue with overthinking" that the authors expect the final version to shake off.

The post frames these releases against a wider licensing split. Google and Meta have moved their open releases to Apache 2.0, while Chinese frontier labs have moved the other way: Kimi K3 "comes with a license which requires commercial agreements for those who run inference or fine-tuning services," and MiniMax M3 "requires agreements above a revenue threshold and has prohibited use cases." The specific revenue numbers for those two are not published in the post.

Shared on Bluesky by 1 AI expert