Research3h ago
Z.ai Releases High-Efficiency GLM-5.3 Model
The 750B parameter GLM-5.3 model outperforms Kimi K3 and rivals Claude Fable 5 on agentic coding benchmarks. Z.ai achieves these results with one-third the parameters of its closest competitor. Open weights arrive on Hugging Face in two weeks. This efficiency proves that scaling post-training can offset raw parameter counts for specialized coding tasks.