GLM-5.3 is here with advanced cyber capabilities — and reportedly already found a 'serious vulnerability' in Cursor

The Chinese AI startup Z.ai has released GLM-5.3, a language model with substantial gains in long-horizon coding and advanced cybersecurity capabilities, which has already found a potentially serious vulnerability in Cursor. GLM-5.3 builds on the 743-billion-parameter-scale base model of GLM-5.2, with improvements coming from scaling post-training across more environments and tasks. The model achieves sizable generation-over-generation improvements on various benchmarks, including Terminal-Bench 3.0, DeepSWE v1.1, and AutomationBench. The practical implication for engineers building AI systems is that GLM-5.3 demonstrates the potential for significant improvements in language models through post-training scaling, rather than relying on expensive pretraining cycles.

⚡ Key Takeaways

• GLM-5.3 achieves a score of 28.3 on Terminal-Bench 3.0, 66.9 on DeepSWE v1.1, and 48.2 on AutomationBench.
• The model uses a 743-billion-parameter-scale base model, with improvements coming from scaling post-training across more environments and tasks.
• GLM-5.3 has found a potentially serious vulnerability in Cursor, demonstrating its advanced cybersecurity capabilities.
• The model is available initially through the GLM Coding Plan and ZCode coding environment, with API access and open weights coming later.
• Z.ai is introducing controls around some of the model's more advanced capabilities, including a "trusted access" approach for sensitive functionality.

The release of GLM-5.3 has significant implications for engineers building AI systems, as it demonstrates the potential for significant improvements in language models through post-training scaling. This approach can be more efficient and cost-effective than relying on expensive pretraining cycles.

✅ Practical Steps

1. Evaluate GLM-5.3's performance on specific tasks and benchmarks to determine its suitability for your use case.
2. Consider using GLM-5.3's advanced cybersecurity capabilities to identify potential vulnerabilities in your systems.
3. Explore the use of post-training scaling to improve the performance of your own language models.

Want the full story? Read the original article.