GLM-5.3 — De-Risked NVFP4
The complete 256-expert GLM-5.3 NVFP4 package with a documented 8× NVIDIA B200 target.
Buy model weights ↓Blackfrost uses The Void method and data generated from reduced-refusal frontier teachers to post-train model behavior. De-Risked does not mean risk-free, refusal-free, universally safe, or appropriate for unauthorized activity.
Requires the Z.AI copyright and permission notice with copies or substantial portions. MaaS operators and affiliates above the stated $10B revenue threshold must complete Z.AI security review before commercial use. Read the official upstream license ↗
Model overview
This catalog edition is derived from Prior Blackfrost GLM-5.3 BF16 master. A one-time purchase licenses the model weights and supporting artifacts in the identified Blackfrost gated repository. The SKU, package description, and verified repository revision below identify what is being offered; upstream rights and notices remain in force.
Release identity and selection guidance
Technical specifications
What the model-weight purchase includes
- Access for the purchaser’s Hugging Face account to the exact gated repository named above
- The release’s model weights, configuration, tokenizer files when present, metadata, and repository documentation
- The SKU, package profile, and verified repository revision stated on this page
- A one-time commercial license subject to the model-specific upstream license and notices
- Delivery support for repository-access issues
What is not included with the weights
- Shadow CLI, an evaluation harness, and other Community or runtime tooling
- An inference API, cloud account, GPU compute, storage, networking, or other infrastructure
- Hosting, installation, migration, application integration, or workload-specific validation
- Custom tuning, post-training, evaluation programs, conversion, or additional quantization
- Ongoing operations, monitoring, incident response, an SLA, or managed support
- Other formats, model families, SKUs, or future catalog releases
Need an operational endpoint? Scalable model hosting is available as a separately scoped service.
Purchase and delivery
- 1Confirm the exact release
Check the edition role, weight package, footprint, runtime notes, upstream license, and deployment boundary before purchase.
- 2Pay securely on Stripe
Enter the Hugging Face username that should receive the weights and complete the one-time payment.
- 3Receive repository access
A signature-verified Stripe event grants the entered account access to the repository above. Purchase confirmation is never inferred from the return URL.
Model access is intended for legitimate research and authorized deployment. Validate model behavior, safety, runtime compatibility, infrastructure capacity, and operational controls against your own workload. Listed configurations are references, not performance guarantees.