Zhilu up 40%, MiniMax up 25%, chip sector also soared! AI concept stocks exploded in Hong Kong

``` Hong Kong AI concept stocks collectively surged on February 12, driven by the intensive release of domestic large language models during the Spring Festival period, which boosted the overall strength of related sectors. Zhipu rose over 40% after the company announced an increase in its AI programming subscription package prices and officially launched its new flagship model GLM-5 the previous day. MINIMAX-WP, another major large model stock, jumped over 21%. On the same day, MiniMax released the latest programming model M2.5, claiming it to be the world's first production-level model natively designed for Agent scenarios. Both companies focus on programming and agent capabilities as their core upgrade directions. The chip sector also saw linked gains. "Chinese GPU Four Little Dragons" saw TianShu Zhixin's afternoon rally expand to 25%, Biren Technology nearly 10%. GigaDevice climbed over 8%. A continued surge in AI computing power demand is raising expectations of benefits for related hardware manufacturers. Behind this rally is the concentration of new launches by domestic large language model providers during the Spring Festival window. Following DeepSeek’s earlier new model release, Alibaba's Qwen 3.5, ByteDance’s SeeDance 2.0, and other products have been unveiled one after another in recent days, marking the industry’s competition entering a white-hot stage. ## Zhipu GLM-5: Parameter Scale Doubled The domestic large model track entered an intensive release phase during the Spring Festival window. DeepSeek launched a new model earlier, with Alibaba Qwen 3.5 and ByteDance SeeDance 2.0 also appearing recently. Multiple companies have simultaneously chosen this time to launch new products, indicating that industry competition continues to heat up. According to a previous Wallstreetcn article, **Zhipu’s GLM-5, launched on February 11, expanded its parameter scale from the previous generation's 355B to 744B, and increased active parameters from 32B to 40B.** Pre-training data grew from 23T to 28.5T. Zhipu confirmed that the mysterious model "Pony Alpha," which previously topped the popularity charts on the global model service platform OpenRouter, is GLM-5. This model introduced DeepSeek's sparse attention mechanism for the first time, reducing deployment costs and improving token utilization efficiency while maintaining long-text processing capability. Architecturally, GLM-5 consists of 78 hidden layers, integrates 256 expert modules with 8 activated at a time, about 44B active parameters, 5.9% sparsity, and supports up to 202K tokens in the context window. Internal assessments show that GLM-5's average performance in programming scenarios such as frontend, backend, and long-range tasks is over 20% better than the previous generation, with real programming experience approaching the level of Claude Opus 4.5. According to Shanghai Securities News, a Shanghai-based AI practitioner analyzed that while domestic large models previously “competed” on who offered the lowest price, Zhipu’s substantial price increase this time **shows a significant improvement in the technical capability and market competitiveness of domestic models.** ## MiniMax Targets International Top Level MiniMax M2.5 is positioned as the world’s first production-grade model natively designed for Agent scenarios, **with its programming and agent abilities directly benchmarked against Claude Opus 4.6.** The model supports full-stack programming development for PC, app, and cross-platform applications, especially leading in core office productivity scenarios such as advanced Excel operations, deep research, and PPT. With only 10B active parameters, the M2.5 model stands out for its low memory usage and inference efficiency, supporting an ultra-high throughput of 100 TPS and inference speeds far surpassing international top models. This rapid iteration follows just over a month since the previous version 2.2’s release. ## The Crowded “Spring Festival Release Season” The Spring Festival of 2026 is no longer just a consumer carnival but has evolved into a race full of speed and passion among China’s AI giants to capture the “mobile end entry point.” According to a latest research report released by JP Morgan on February 11, China’s Internet and AI industries are experiencing the most intensive wave of flagship model releases in history. This is no longer a solo performance by a single model, but a game of musical chairs to see who can most quickly turn “technology spillover” into a “consumer blockbuster.” > **ByteDance** was the first to make a move, releasing a “package” of three models: Seedance 2.0 (video), Seedream 5.0 (image), and Doubao 2.0. Among them, Seedance 2.0 has already shown "blockbuster" potential. > > **Alibaba** is also unwilling to lag behind, reportedly preparing to launch Qwen 3.5 in mid-February, along with a 3 billion yuan incentive plan to drive user acquisition. > > **DeepSeek’s** V4 version is reportedly targeted for mid-February, focusing on improvements in coding and ultra-long prompt handling. As of February 11, there were reports that DeepSeek has updated its new model, supporting up to 1M (million) token context length. Risk Disclosure and Disclaimer Markets are risky and investments should be made with caution. This article does not constitute personal investment advice and does not take into account individual users’ special investment objectives, financial situations, or needs. Users should consider whether any opinions, views or conclusions cited herein are suitable for their particular circumstances. Investment is at your own risk. ```