Zhipu’s GLM-5.3 Tops US Rivals in Cybersecurity Benchmark
Chinese artificial-intelligence developer Zhipu has introduced GLM-5.3, a large language model focused on coding and reasoning, and said it plans to release the model weights. The move could broaden access to advanced software-analysis capabilities, including tools that help researchers identify vulnerabilities, while intensifying debate over the security benefits and potential misuse of openly available AI models.
GLM-5.3 scored 84.5% on CyberGym, a benchmark designed to test vulnerability-discovery performance, according to results released by Zhipu. The company said the score exceeded those of flagship models from Anthropic and OpenAI. The headline result points to stronger Chinese competition in specialized coding tasks, though comparisons depend on testing conditions, tool access and how successful vulnerability discovery is measured.
All Coverage
1 original reportsThe Backstory
The history behind this eventZ.ai's GLM-5.2 Tops Open-Weight Model Rankings, Beating DeepSeek
Zhipu, also known as Z.ai, is a Beijing-based AI company that has long developed the GLM family of large language models. Open model weights allow developers to deploy and fine-tune models themselves, reducing reliance on closed platforms. Artificial Analysis' Intelligence Index combines multiple reasoning and knowledge benchmarks, and GLM-5.2's top ranking reflects the continued rise of China's open-weight models.
Zhipu released GLM-5.2 under the MIT License on June 13, 2026. The model has 744 billion total parameters, activates 40 billion parameters per computation and supports a context window of 1 million tokens. It scored 51 on the Artificial Analysis Intelligence Index v4.1, surpassing DeepSeek V4 Pro's score of 44. Its API charges $1.40 per million input tokens and $4.40 per million output tokens.
Zhipu AI Open-Sources GLM-5.1, Prices Model in Line With Global Rivals
Zhipu AI is one of China’s six leading large-model startups, and its GLM series initially expanded its market presence through open-source releases and competitive pricing. GLM-5.1 focuses on coding and long-horizon tasks, while its pricing has moved closer to Anthropic’s. The shift indicates that Chinese model developers are moving away from competing primarily on low prices and toward challenging global leaders on technical capabilities.
On April 8, 2026, Zhipu AI formally open-sourced GLM-5.1 and raised its API prices by 10%, to about $1.4 per million input tokens and $4.4 per million output tokens. The model ranked third globally and first among open-source models on a composite of three coding benchmarks, and can operate autonomously for as long as eight hours in a single session. Zhipu AI’s Hong Kong-listed shares rose nearly 19% intraday.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →