Z.AI (02513) released its interim results, with revenue of 954 million yuan, a year-on-year increase of 399.7%.

Z.AI (02513) released its interim results for the six months ended June 30, 2026, reporting revenues of 954 million yuan, a year-on-year increase of 399.7%; gross profit of 252 million yuan, a year-on-year increase of 163.7%; R&D expenses of 2.131 billion yuan, a year-on-year increase of 33.64%; adjusted net loss of 1.964 billion yuan, a year-on-year increase of 12.1%; and a loss per share of 4.67 yuan. In the first half of 2026, the company validated the deliverability of large models in programming, progressing from early simple coding recommendations to complete code file generation and eventually realizing an entire engineering project (Agentic Engineering). The company has fully completed this evolution: the model is no longer just a tool, but has the capability to independently accomplish a task. Consequently, the revenue structure has shifted, with the proportion of income from open platform and API services billed by usage rising from 26.3% in the previous year to 86.5%, while revenue from localized model deployment dropped from 73.7% to 13.5%. The main revenue structure transitioned in the first half of 2026, which is essentially an inevitable occurrence following the leap in model capability. Cybersecurity Cowork is an extension of the same capability curve of the model. Software vulnerability discovery no longer relies on memory but on the depth of reasoning and long-range autonomy of the model, which is a natural extension of programming capabilities after surpassing the engineering project threshold. During the reporting period, the company's model achieved a score of 84.5% on CyberGym (surpassing Fable5 and GPT-5.6Sol) and completed 130 tasks on ExploitGym (second only to Fable5 and GPT-5.6Sol). Supporting these two curves is the infrastructure. Since early 2026, the company has fully implemented the All-in-Infra strategy, investing simultaneously in training and reasoning while incorporating domestically produced chips into its main reasoning computing power. The company has achieved large-scale, low-cost reasoning with domestic chips at a scale of 100,000 units, reducing the cost of reasoning per token by 80% compared to the beginning of the year. Amidst a globally tight supply of computing power, this path ensures that the company's cost curve is under its own control. By the end of the reporting period, the total number of enterprises and developer users on the GLM model's MaaS platform exceeded 5.8 million. The total revenue amounted to approximately 954 million yuan (about 142 million USD), a year-on-year growth of nearly 400%. Among this, revenue from open platform and API business was around 825 million yuan (about 123 million USD), reflecting a year-on-year increase of 2,735.7%. In August 2026, the company released GLM-5.3 and GLM-5.3-Flash (Ox-Alpha), and simultaneously opened the GLM Coding Plan subscription to all users, covering Lite, Pro, Max, and team versions. This is one of the earliest large model services in China delivering intelligent coding capabilities on a subscription basis. As of the date of this announcement, the number of enterprises and developer users on the GLM model's MaaS platform exceeded 7.4 million.