GLM-5.3-Flash is an open-weight model from Z.ai that now has two published derivatives. OrcaCyber released abliterated weights on October 27, 2026, and Z.ai's GLM-Legal, built on GLM-5.3-Flash, is in beta.
Z.ai's GLM-5.3-Flash is an open-weight model, and the stories about it are mostly about what other people built on top of it. The earliest of the three, from September 18, 2026, describes the production inference service behind the model: a cluster of more than 100,000 Chinese-made AI accelerators, with most of the optimization work done by an infrastructure agent powered by GLM-5.3. Z.ai put end-to-end service performance at about 3x, with cost per token in line with mainstream NVIDIA GPUs.
In October the model became a base for specialized work. Z.ai announced GLM-Legal on October 13, 2026, tuned on GLM-5.3-Flash for legal tasks and with a harness also optimized for legal work. The company said GLM-Legal improved on its base model and surpassed Claude Fable 5 and GPT-6 Astra on some tests, with comparisons also run against GLM-5.3-Flash and Kimi K3. GLM-Legal is in beta testing, is scheduled to reach Z.ai's AI agent AutoClaw, and is accepting API access applications.
Two weeks later the model's open weights took a different direction. OrcaCyber, the cybersecurity team at the OrcaRouter inference gateway, released weights for an abliterated version of GLM-5.3-Flash on October 27, 2026. The abliterated model relaxes refusal responses for security research and is offered free for 48 hours through the Orca Cyber Tier API, alongside OrcaCyber Zero and OrcaCyber Ultra. OrcaRouter is provided in Japan by FlashLabs.
Written with AI from the cited Yomimono stories below, each itself sourced. See our editorial policy. Every fact links the story it came from.
Structured graph also available as JSON at /public/entities/glm-5-3-flash.
CC BY 4.0.
OrcaCyber, the cybersecurity team at the OrcaRouter inference gateway, released weights for an abliterated version of Z.ai's open-weight GLM-5.3 Flash on October 27, 2026. The model relaxes refusal responses for security research and is offered free for 48 hours through the Orca Cyber Tier API, alongside OrcaCyber Zero and OrcaCyber Ultra. OrcaRouter is provided in Japan by FlashLabs.
Z.ai announced GLM-Legal, a model tuned on GLM-5.3-Flash for legal work. Its harness is also optimized for legal tasks. The company compared benchmark scores against GLM-5.3-Flash, Kimi K3, Claude Fable 5, and GPT-6 Astra, saying GLM-Legal improved on its base model and surpassed Claude Fable 5 and GPT-6 Astra on some tests. GLM-Legal is in beta testing, is scheduled to reach Z.ai's AI agent AutoClaw, and is accepting API access applications.
Z.ai explained in a blog post how it built the inference infrastructure behind the production service for GLM-5.3-Flash, running on a cluster of more than 100,000 Chinese-made AI accelerators. The company says most of the work was carried out by an infrastructure agent powered by GLM-5.3, and that optimizations brought end-to-end service performance to about 3x and cost per token in line with mainstream NVIDIA GPUs.