Zhipu's Tang Jie explains why GLM-5.3 scales post-training, not parameters
智东西 · wechat · 2026-08-19
Zhipu co-founder and chief scientist Tang Jie posted a long essay on X explaining his view of Scaling Law and why GLM-5.3 didn't grow in parameter count.
He argues parameters can't be discussed apart from data size, compute placement and deployment: Kaplan 2020 pushed the field toward huge models (GPT-3, Gopher), Chinchilla 2022 corrected to balanced growth, and the trillion-parameter chase was a collective detour. As inference now dominates lifetime cost, the optimum shifts to deliberately over-trained smaller models (Llama-2-7B); with MoE, total params govern knowledge capacity while active params govern reasoning depth.
GLM-5.3 is a controlled experiment: identical base model, architecture and parameter counts as GLM-5.2, with one month of expanded long-horizon RL environments yielding non-trivial gains. Scaling has multiple knobs—post-training simply had the most headroom this time, and Zhipu may return to pretraining or mid-training next.
More from Models
- Meta's coding-optimized open-weight Muse Spark Open lands on Modal with 1M context — AAAzzam · 2026-10-16
- Google launches Gemini 4 Argon frontier model on Cloud with 10 official demos — Saboo_Shubham_ · 2026-10-16
- RL practitioner: ~10,000 tasks still stump every frontier coding model — max_paperclips · 2026-10-16
- SpecID eval falls: Gemini 3.5 Flash scores 67%, beating its human creator at Zimbabwean species ID — johnowhitaker · 2026-10-16
- Researcher criticizes Claude for lecturing users and refusing critical takes on tech firms — BlancheMinerva · 2026-10-16
- Chart: Meta went 492 days without an open-weight release before Muse Spark Open — demian_ai · 2026-10-16