Zhipu AI's GLM-5.3 boosts coding by 50%, finds 40-year-old vulnerability · 1 source tracked

Zhipu AI has released GLM-5.3, an updated version of its large language model, with significant improvements in programming and security capabilities. The model's coding performance has reportedly increased by 50%, nearing top-tier models like Claude Fable 5 and even surpassing Claude Opus 4.8 in certain benchmarks. GLM-5.3 also excels in identifying vulnerabilities, having discovered over 2,400 bugs, including nearly 1,100 high-risk ones, some of which had been dormant for decades. These advancements are attributed to Zhipu AI's refined post-training techniques, including IndexShare, SAO, and the Slime framework, which optimize architecture, reinforcement learning, and training infrastructure. AI

IMPACT Sets a new benchmark for open-source models in coding and security, potentially accelerating enterprise adoption of AI for software development and cybersecurity.

RANK_REASON The cluster reports on the release of a new version of a large language model by a frontier lab (Zhipu AI). [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 雷峰网 (Leiphone) →

• Claude Fable 5
• Claude Opus 4.8
• DeepSeek
• GLM-5.3
• GPT-5.6 Sol
• Hugging Face
• Kimi K3
• Meta
• Mythos 5
• OpenAI
• Zhipu AI

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →