Z.ai releases GLM-5.3's weights under a new license requiring companies with $10B+ in revenue over 12 months to pass Z.ai's security review to host the model
Z.ai put the GLM-5.3 weights on Hugging Face, but dropped the MIT license. Providers above $10 billion in revenue now need a security review.
The New StackFrederic Lardinois
Context & Ripple Effects
Z.ai positioned GLM-5.3 as a post-training upgrade to the same base model used by GLM-5.2, after releasing GLM-5.2 under MIT terms for agentic coding and long-horizon work. The weights release fulfills that earlier distribution plan, but replaces the permissive license model with a revenue-based hosting condition.
The change matters because Z.ai is separating downloadability from unrestricted commercial serving. Public reaction focused on how the affiliate language applies and on the absence of model-card, safety-testing and third-party-evaluation materials, concerns that matter more for a model Z.ai markets for coding and cyber defense.
First-order effects
Hosting providers with more than $10 billion in 12-month revenue must obtain Z.ai's security review before serving GLM-5.3, while hosts below that threshold are not subject to that stated requirement.
Z.ai retains a direct approval point over a portion of commercial deployment even though the model weights are publicly downloadable.
Second-order effects
Large cloud and model-serving companies face an extra licensing and compliance step relative to smaller hosts, making the terms of access part of their model-selection calculus.
The revenue threshold gives Z.ai leverage over where high-capacity commercial hosting occurs, rather than leaving distribution entirely to downstream infrastructure providers.
Third-order effects
If other open-weight developers adopt comparable terms, open weights may increasingly function as distribution channels paired with provider-specific control over high-scale deployment.
Security review requirements can make model licensing a competitive control plane, with the practical openness of a release differing by the size and identity of the host.
The trend: Open-weight AI distribution is evolving from permissive publication toward downloadable models with selective controls on commercial-scale hosting.
GLM-5.3 is now open-weight. Our most capable model for agentic coding and cyber defense is now available to download, run, and customize. Weights: https://huggingface.co/... Tech blog: https://z.ai/...
GLM 5.3 Flash performs at GLM 5.3 level in Blender for 17x cheaper! We gave both models a live Blender over MCP and one prompt: a 2,800 sq ft duplex penthouse, double-height living room, mezzanine, floating stair, curtain wall, terrace, furnished, real PBR materials Outputs:
would mistral be able to serve GLM 5.3 legally? what counts as “affiliates” here? (like does ASML count?) also the 10B USD figure here is very high, other licenses like Kimi K3 have a threshold of 20M USD, so 3 orders of magnitude less
I'm happy that more Chinese companies are experimenting with the Kimi License. This one is $10 Billion USD vs Kimi $20 Million USD. Aggregate revenue over 12 months rev. This gives US companies a template to copy.
afaict no model card, no safety testing, no third party evals There is a safety case that could be made for open weight releasing a model this capable (cyber defense vs offense balance etc), but they didn't even really bother to try to make it
GLM-5.3 is a good model, and as the open weights models get better and better it becomes increasingly important that they actually publish model cards, do red teaming, etc. Since you can break the guardrails with any open model, we need a sense of what the risks are as well.
GLM 5.3 is now available in Perplexity Computer. Built for long-context, multimodal agent workloads, it beat GLM 5.2 on WANDR, our benchmark for large-scale, evidence-backed research.
GLM-5.3 is now available for download, local deployment, fine-tuning, and commercial use under the GLM-5.3 License. Given the model's advanced cybersecurity capabilities, we conducted two additional weeks of comprehensive safety evaluations before releasing the weights. Under
GLM-5.3 matches the frontier with just 753B total parameters. It is the same architecture as GLM-5.2. This model is a triumph of post-training. There is incredible ROI to unlock by applying more compute to RL across domains.