CICC has issued a research report, keeping its profit forecasts for KNOWLEDGE ATLAS (ASX: 02513) largely unchanged. Due to the company's rapid model iteration exceeding both market and the firm's expectations, CICC expresses optimism regarding the accelerated release of its forward-looking API. Consequently, the target price has been raised by 39% to HK$1250, implying a 14% upside from the current share price. The outperform industry rating is maintained. The firm's primary views are outlined below.

Recent Company Developments

On June 13th, the company's latest model, GLM-5.2, was made available to all users of the GLM Coding Plan, covering Lite/Pro/Max/Team editions. Simultaneously, the company announced that the GLM-5.2 API will officially launch and be open-sourced in the third week of June, following the MIT license. As the company's newest and most intelligent open-source model, GLM-5.2 continues to focus on the coding domain. It has extended its context window to 1M tokens, demonstrating improved capabilities in long-context tasks. The model has received excellent feedback in numerous developer evaluations and is the first domestic model to achieve real-world performance comparable to Claude-Opus 4.6, with programming capabilities firmly leading among domestic models and matching the top tier of international models.

Continued Focus on Coding, Significant Improvements in Front/Back-End Programming

Based on the firm's and numerous developers' evaluations, GLM-5.2's overall programming capabilities show clear advancement compared to GLM-5.1. In its strong suit of backend tasks, it can now match Claude Opus 4.6 in several evaluation benchmarks. The aesthetics, precision, and controllability in frontend tasks have also seen notable improvement. Its comprehensive programming ability continues to lead domestic models, marking it as the first domestic model to achieve real-world performance on par with Opus 4.6, gradually widening the gap with other domestic competitors.

Context Window Extended to 1M Tokens, Enhanced Long-Context Task Performance

The firm believes GLM-5.2 incorporates numerous optimizations at the model architecture level. The extension of the context length from 200K to 1M tokens addresses the previous shortcomings of GLM-5.1 in long-context tasks, significantly boosting its usability in real-world, complex programming scenarios. The model's ability to handle complex, multi-turn agent tasks has also improved markedly.

Proactively Addressing Compute Bottlenecks to Improve User Experience and Supply

Due to the model's outstanding capabilities, overall demand exceeds supply, leading to constraints from limited computational resources. GLM-5.2 may also exhibit longer response times during use, which the firm attributes primarily to high concurrency and compute limitations. CICC anticipates the company will actively address these compute challenges through various means, effectively alleviating supply-side bottlenecks.

Risk Factors

Key risks include slower-than-expected model iteration and user feedback falling short of expectations.