GLM 5.3 release
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Z.ai released GLM-5.3 on Aug 14, 2026, claiming a 50% coding improvement over GLM-5.2. The open-weights model's cyber capabilities grew faster than expected, delaying public release.
Background
Z.ai (Zhipu AI) released GLM-5.3 on August 14, 2026, as an upgrade to its GLM-5.2 model. The company says all gains come from scaling post-training, not a new base model. The model shows significant improvements in coding and cybersecurity benchmarks, but its advanced cyber capabilities have raised safety concerns, leading to a two-week delay in releasing the weights.
How it unfolded
- Aug 14, 2026Z.ai released GLM-5.3 through its GLM Coding Plan and ZCode, claiming it is the strongest open-weights coding model.
- Aug 14, 2026Z.ai announced that GLM-5.3's weights would be released in about two weeks after safety evaluations.
- recentlyZ.ai announced a tiered access program for the most sensitive cyber functions, available only to verified users.
Who’s saying what
- Official
- Z.ai says GLM-5.3 is the most capable open-weights coding model, with a 50% improvement over GLM-5.2 on its internal benchmark.
- Caution
- Independent evaluators have not yet reproduced Z.ai's benchmark numbers, so the true performance remains unverified.
- Support
- Z.ai frames the cyber capabilities as a benefit for defenders, offering a program for open-source maintainers to scan their repositories for bugs.
Still unverified
Z.ai's claim of a 50% coding improvement and 'strongest open-weights coding model' are based on its own evaluations; independent benchmarks are not yet available. The reported discovery of 2,436 vulnerabilities across 269 projects is from Z.ai's own security work and has not been independently verified.