Chinese artificial intelligence startup Zhipu AI launched its next-generation foundation model GLM-5.3 on Friday, claiming coding and AI agent capabilities approaching Claude Fable 5.

According to the company, the model pushes the boundaries of China's open-source AI models beyond coding and reasoning, with cybersecurity emerging as a new battleground as increasingly capable AI agents learn to identify vulnerabilities, operate software and execute complex tasks with less human intervention.

GLM-5.3 has 743 billion parameters, with much of its performance improvement coming from post-training rather than simply expanding model size, Zhipu said. The company said the model made significant gains in coding and complex engineering execution.

According to a technical report released by Zhipu, GLM-5.3 scored 28.3 on Terminal-Bench 3.0, which the company said was the highest result among open-source models and ahead of Moonshot AI's Kimi K3.

The model also scored 66.9 on DeepSWE 1.1, another benchmark focused on software engineering capabilities.

Zhipu said GLM-5.3 ranked as the strongest open-source model across several mainstream benchmarks, with its coding and agent capabilities approaching those of Claude Fable 5. It also said the model's practical coding performance surpassed other Chinese models.