According to CoinMeta, during Zhispu's Coding turnaround, DeepSeek forced it to change direction, and users helped to choose a new path for it. Based on a review by 36Kr, after the DeepSeek R1 became extremely popular during the Spring Festival of 2025, Zhispu's business that relied on enterprise customization was impacted, with nearly 30% of customers turning to DeepSeek. To retain these customers, Zhispu temporarily offered discounts. In May, Zhispu held a key strategic meeting and decided to integrate three types of capabilities: reasoning, coding, and agentic into a single large model, delaying the release of GLM-4.5 which was originally scheduled for April. User feedback indicated that "improving R&D efficiency" was a major demand. After GLM-4.5 went live, more and more enterprises requested to integrate it into their programmers' workflows, and coding quickly became the most well-received feature. Subsequently, Zhispu scaled back its operations, reducing the team responsible for coding and text-related tasks to about 100 people, and the multi-modal training team was cut down to 10 members. After GLM-4.5, Zhispu launched coding and plan, as well as an enterprise version. coding became one of the important capabilities of the GLM series, and revenue soared from $500 million in May to $100 million.