New Models
AI Minute Newsroom
2026-08-17
Zhipu did not build a new model. It kept training the old one — and says coding got 50 percent better.
Zhipu released GLM-5.3 on 14 August. The unusual part is what it is not: the base model is unchanged from GLM-5.2 — the same roughly 744-billion-parameter mixture-of-experts design with about 40 billion parameters active per token. Every claimed gain comes from post-training: a longer training run, tens of times more long-context task environments, and more varied kinds of them. Zhipu says coding ability improved about 50 percent over GLM-5.2, ranks first among open-weight models on Terminal Bench 3.0 and Agents' Last Exam, and scores 84.5 percent on the CyberGym security benchmark. On the company's own Code Bench it puts GLM-5.3 at 31.4 percent against Claude Opus 4.8 at 29.5 percent, using about 50,000 output tokens where Opus used about 120,000. Most of these numbers are Zhipu's own evaluations. The weights are not out yet: the company says it will publish them roughly two weeks after release, once security hardening is finished.
Why it mattersFor two years the recipe for a better model has been a bigger, freshly trained base. This is a claim that you can leave the base alone and get a large jump from what you train it on afterwards — which is far cheaper and is something a small lab could copy. Treat the size of the jump with caution until the weights are public: almost every number here was produced and reported by the company that sells the model, and nobody outside it has checked them yet.
✓ Verified · 3 sources
Read in the app — free, in 9 languages
Related stories
OpenRouter dropped three AI providers after a bait-prompt leak test.
2026-11-22Tencent released the weights of its flagship Hy4 model for download.
2026-11-21METR says it can rarely find humans who beat top AI at coding tasks.
2026-11-21Thinking Machines releases Inkling 2 Mini as an open laptop model.
2026-11-21Cursor tells Congress it may drop Chinese AI models over risk.
2026-11-20