CICC has released a research report maintaining an "Outperform" rating on KNOWLEDGE ATLAS (02513). The company has launched its foundational model, GLM-5-Turbo, which is deeply optimized for the OpenClaw lobster scenario. CICC has kept its profit forecast for the company unchanged. Due to the rapid iteration of the company's model capabilities, the institution is optimistic about the accelerated future release of its API, leading to a 16% increase in the target price to HKD 800 (corresponding to a 2028 P/S ratio of 47x and a discount rate of 7%). This implies a potential upside of 32% from the current share price. CICC's main points are as follows:
Recent Company Developments: On March 16, the company launched the GLM-5-Turbo model, which is deeply optimized for the OpenClaw lobster scenario, focusing on resolving issues users face with OpenClaw, such as stability, security, and enterprise usability. CICC believes the popularity of OpenClaw has spurred incremental demand for Agent deployment from both individual and enterprise users. The launch of GLM-5-Turbo is expected to accelerate the penetration of Agent capabilities across various industries, serving as a new growth engine—besides Coding—driving the rapid scaling of the company's API.
As a native lobster model, GLM-5-Turbo's core capabilities focus on being Agentic. The company systematically constructed multiple types of lobster task scenarios, from training data structure to optimization target design, all centered around real Agent workflows. This enhances the model's executability in complex, dynamic, and long-chain tasks, with key improvements in core capabilities such as tool invocation (external tools + various Skills), instruction following, timed and persistent tasks, and high-throughput long-chain execution (stable and fast). CICC believes that as a native lobster model, GLM-5-Turbo integrates more smoothly with various scenarios in the Agent field, improving the user experience.
Focus on Real User Experience; GLM-5-Turbo Receives Widespread Acclaim: In the company's released lobster scenario benchmark, ZClawBench, GLM-5-Turbo led in several key tasks compared to many mainstream domestic and international models. In real user testing, GLM-5-Turbo received widespread positive feedback during its internal testing phase from internet companies like Alibaba, ByteDance's Coze, and Meituan, particularly regarding long-range task execution, instruction following, and stability.
API Price Increased by 20%, Launch of Lobster Packages to Cover Diverse Customers: The price of GLM-5-Turbo has been increased by over 20% compared to GLM-5, with input/output prices set at $1 and $4 per million tokens, respectively. Additionally, the company has launched lobster packages based on GLM-5-Turbo, including a Personal edition and a Team edition. The Claw Experience monthly card is priced at RMB 39 per month, covering 35 million tokens, while the Advanced monthly card costs RMB 99 per month, covering 100 million tokens. CICC believes that as model intelligence improves and Agent scenarios penetrate further, more users will be willing to pay for intelligence. Simultaneously, the price increase provides more room for gross margin improvement.
Risk warnings include slower-than-expected model iteration and user feedback falling short of expectations.
Comments