CLSA has released a research report stating that since the official launch of KNOWLEDGE ATLAS (ASX: 02513)'s GLM-5.2 API on June 17, substantial evidence has emerged indicating that KNOWLEDGE ATLAS AI has ascended to global state-of-the-art (SOTA) levels in the field of code generation. Its performance is now comparable to that of Opus 4.6 to 4.8, despite its model size being significantly smaller than its competitors. The global community also holds optimistic expectations, believing China has the potential to achieve Fable-grade AI capabilities by the first quarter of 2027, or even earlier. The firm assigns KNOWLEDGE ATLAS a "Hold" rating.

CLSA views this development as a landmark moment signifying global formal recognition of China's post-training capabilities. It is also seen as strong evidence of KNOWLEDGE ATLAS AI's data flywheel effect, reflecting a major victory for the company in the field of reinforcement learning. This is described as a true "GLM moment," occurring approximately 16 months after DeepSeek's breakthroughs in pre-training and infrastructure.

The report notes that GLM-5.2 has significantly enhanced the appeal of KNOWLEDGE ATLAS AI's API services. However, it is believed that the company's near-term Annual Recurring Revenue (ARR) growth will still largely depend on the availability of computing power resources. The official pricing aligns with that of GLM-5.1 for inputs above 32k, at approximately RMB 8 per million input tokens and RMB 28 per million output tokens. However, if enterprises extensively utilize the 1 million token context window, this memory-intensive application could lead to a significant increase in per-token inference costs. Therefore, more data is required to assess the trend in profit margins.