Z.ai launches GLM-5.3, built for coding and cyber defense
- Source
- x.com
- Date
GLM-5.3 is coming. focusing on coding. https://t.co/g10V8eJIwq
Context
Z.ai's teaser tweet for GLM-5.3 says only that the model is 'coming' and focused on coding, but the company's full announcement explains the release is a controlled experiment: GLM-5.3 uses the identical 743-billion-parameter base as GLM-5.2, with all reported gains coming from roughly a month of additional post-training rather than a new run. On Terminal-Bench 3.0, a test of an completing real terminal-based coding tasks, Z.ai reports GLM-5.3 jumping from GLM-5.2's 4.6 percent to 28.3 percent.
The 'cyber defense' framing refers to CyberGym, a testing whether a model can identify and validate real vulnerabilities from source code; Z.ai reports GLM-5.3 scoring 84.5 percent there, up from GLM-5.2's 77.2 percent. These are Z.ai's own reported benchmark numbers rather than independently reproduced ones. GLM-5.3 shipped through Z.ai's GLM Coding Plan and ZCode to existing subscribers first, with general API access following several days later at GLM-5.2 pricing.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
- pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…