资本市场对中国上市大模型公司的估值逻辑正在发生深刻转变。曾经能直接驱动股价上涨的技术排名和最先进模型（SOTA）地位，如今正在失去估值支撑力。过去12个月内出现了107个显著模型发布，平均约3天发布一次，前沿模型的发布周期已缩短至1至2周[1]。这种加速迭代打破了单一模型能维持长期领先优势的可能性。
市场的这一转变在股价波动中得到了充分体现。7月17日，智谱发布Kimi K3后，其股价在两个交易日内下跌42.47%，MiniMax同期跌幅约24.57%[1]。相比之下，6月GLM-5.2发布使智谱评分进入全球前三后，智谱首个交易日上涨32.82%，市值曾突破万亿港元[1]——这种巨大的反差说明市场已不再为模型排名的单纯改善而持续追捧。7月31日DeepSeek V4 Flash和MiniMax H3发布后，三家公司股价虽有反弹（智谱上涨14.56%、MiniMax上涨13.15%、阿里收涨4.73%），但涨幅明显小于早前的峰值[1]。
资本市场现在更看重的是公司的实际商业化能力。迭代速度、收入增速和客户留存等指标已成为估值的核心考量因素，而非单纯的技术排名[1]。这一转变也反映在机构评级中。J.P.Morgan将智谱的目标价从2400港元下调至1600港元，降幅约33%[1]。随着大模型性能竞争逐渐常态化，投资者正在将目光转向那些能够将技术优势转化为商业价值的企业。
The valuation logic for Chinese listed large language model companies is undergoing a significant shift, with market participants increasingly discounting the importance of achieving state-of-the-art (SOTA) model rankings in favor of commercial metrics.[1]
Over the past twelve months, approximately 107 significant model releases have occurred at an average pace of one every three days, with frontier models now being released at intervals of one to two weeks.[1] This accelerated release cycle has fundamentally altered investor calculations. When GLM-5.2 was released in June, it achieved a score of 51 on the AA Intelligence Index, securing entry into the global top three, and Zhipu's stock surged 32.82% on the first trading day, with market capitalization briefly exceeding one trillion Hong Kong dollars.[1] However, the sustainability of such gains has proven limited. Following the release of Kimi K3 on July 17, Zhipu's stock fell 42.47% over two consecutive trading days, while MiniMax declined approximately 24.57%.[1]
The market's recalibration became evident when subsequent model releases on July 31 triggered a reversal. DeepSeek V4 Flash, identified as the most cost-effective model for single-task applications at 105 times cheaper than Anthropic's Claude 5, was released alongside MiniMax H3.[1] This time, Zhipu's stock rose 14.56%, MiniMax climbed 13.15%, and Alibaba gained 4.73%.[1] When Alibaba released Qwen 3.8-Max in its official version on August 3, featuring 2.4 trillion total parameters and 95 billion activated parameters, the stock gained 6.75%.[1] Rather than rewarding technical supremacy alone, these movements suggest investors are now prioritizing iteration speed, revenue growth, and customer retention as indicators of sustainable business performance.[1] This reorientation is reflected in analyst adjustments: J.P. Morgan downgraded Zhipu's price target from 2,400 Hong Kong dollars to 1,600 Hong Kong dollars, representing a 33% reduction.[1]
