Zhipu GLM-5.3 Released: Significant Gains in Coding and Agent Capabilities
AI寒武纪 · wechat · 2026-08-15
Zhipu released the GLM-5.3 model, fine-tuned from a 743B base, claiming top-tier performance in coding and agent capabilities within China.
Benchmark Performance:
- Terminal-Bench 3.0: Score surged from 4.6 to 28.3.
- DeepSWE v1.1: Improved from 46.2 to 66.9.
- Z.ai CodeBench: Achieved 31.4% accuracy, surpassing Claude Opus, with token usage cut by half.
Real-world Tests:
- Physics Simulation: GLM-5.3 generated a high-precision Mars landing simulation purely through reasoning, outperforming vision-enabled competitors.
- GitHub Analytics Panel: Built a full-stack project with user search, trending lists, AI analysis, and strict engineering standards (TS, unit tests, caching) in just one hour with zero bugs, showing significant efficiency gains over the previous version.
Cybersecurity Capabilities:
- Excels in code review and vulnerability verification (CyberGym score: 84.5%, slightly beating Claude and GPT).
- Red team exercises with universities and security firms uncovered 2,436 real-world vulnerabilities (including 1,097 medium-to-high risk ones), some dormant for over 45 years.
Availability:
Integrated into ZCode and AutoClaw; API coming soon, with full weights expected to be open-sourced within two weeks.
More from Models
- Google says teams across the company are using Gemini 4 Argon on its hardest engineering problems — CurieuxExplorer · 2026-10-15
- Perplexity's new embedding models: index with a 9B model, search with 0.6B — antoine_chaffin · 2026-10-15
- 797GB of n-gram counts in CPU RAM let 1B models match 57% more compute — PangWeiKoh · 2026-10-15
- Leak: Moonshot's Kimi flagship 'x4' done with pre-training, K3.1 in internal testing — lyraxana · 2026-10-15
- Unsloth ships GGUF quants for Nemotron 3.5: Bolt-74B runs in 24GB RAM — danielhanchen · 2026-10-15
- NVIDIA ships two open Nemotron 3.5 Bolt models: a 74B-A9B MoE and a mobile-friendly E4B — HankYeomans · 2026-10-15