CICC 2026 H2 Outlook: Upward Cycle Trend Unchanged in Midstream of New Energy Vehicles, Fundamentals Resonate with New Technologies 中金2026下半年展望：新能源车中游上行周期趋势不改，基本面与新技术共振

Dubbed "the biggest overhaul ever," the core change is liberating the model's capabilities from a single dialogue box, enabling it to call plugins, browse the web, analyze data, and even generate images. From a product perspective, this feels like an inevitable evolution. Users are tired of jumping between countless vertical AI applications, weary of having to precisely describe tasks in natural language only to receive a dry, text-based answer. What they want is to simply "throw" a task at the AI and get a result—whether it’s travel planning, data analysis, or a creative sketch. The super app concept is essentially using AI to redefine the very idea of an "operating system."

Yet, buried here is a profound irony: While everyone else chants "model supremacy," believing bigger parameters and longer context windows are everything, OpenAI is frantically playing catch-up—on product experience, on tool ecosystems, on interaction design. This tacitly admits a reality: As a standalone product, even a powerful pure large language model like GPT-4 has a visible ceiling. It is smart but clumsy; it is knowledgeable but cannot act proactively. It needs a shell, needs eyes and limbs.

This reminds me of the predicament faced by many domestic AI assistants in China. Most are still stuck in the "smarter Siri" phase, competing over who sounds more human or who responds faster. But users will quickly realize these flashy conversations cannot solve any complex problem. If you ask an AI assistant to book a flight, it can give you a textual description, but you still have to open the app yourself to complete the task. ChatGPT’s pivot is essentially saying: Stop just chatting—get the job done.

However, is OpenAI’s path necessarily the right one? Shoving all functionality into a single app risks creating another "super bloatware." Historically, from Microsoft’s Office to various domestic "suite packages," this ambition to solve all needs in one place often ends in complexity, bloat, and innovation stagnation. AI’s strength lies in its modularity—a powerful core model can drive countless lightweight, specialized applications. Forcing a monolithic approach may go against the natural trend of technological diversification.

What’s more intriguing is the news about Xiaomi. As ChatGPT strives to "expand outward," does Xiaomi’s self-developed large model and AI assistant face the same choice? Should it continue to deeply integrate into core scenarios like smartphones and smart homes, becoming an "ambient intelligence that is everywhere," or attempt to build a general-purpose, omnipotent super assistant? The two paths are radically different. Xiaomi’s advantage actually lies in its hardware ecosystem, which might be a more realistic moat than building a universal chat interface.

Ultimately, ChatGPT’s overhaul is a weather vane, marking AI competition’s shift from a "model arena" to a "product implementation phase." Whoever can seamlessly and unobtrusively weave AI capabilities into users’ concrete task flows—rather than forcing users to learn new interaction methods—will likely win the next round. In this sense, opportunities may be greater for companies like Xiaomi, because they are closer to users’ "concrete lives." Meanwhile, OpenAI faces the thrilling leap from its "technical pedestal" into the "application red ocean." This transformation—if successful, it could elevate them to legend; if not, they risk becoming another giant trapped by their own creation.

ChatGPT要从聊天框里爬出来了，这是好事，但也暴露了一个根本性的问题：我们到底需要一个怎样的AI助手？OpenAI显然给出了它的答案——一个吞噬一切的“超级应用”。

所谓“史上最大改版”，核心就是把模型能力从一个对话框里解放出来，让它能调用插件、浏览网页、分析数据、甚至生成图片。从产品角度看，这简直是必然的进化。用户厌倦了在无数个垂直AI应用里跳来跳去，厌倦了用自然语言精确描述任务，然后得到一段干巴巴的文字回答。他们需要的是直接把任务“扔”给AI，然后得到一个结果，无论是行程规划、数据分析还是创意草图。超级应用，本质上是在用AI重新定义“操作系统”的概念。

但这里面藏着一个巨大的讽刺：当所有人都在高呼“模型为王”，认为更强的参数和更长的上下文就是一切时，OpenAI却在拼命补课——补产品体验、补工具生态、补交互设计的短板。这等于默认了一个事实：纯粹的大语言模型，哪怕强大如GPT-4，其作为独立产品的价值天花板已经肉眼可见。它聪明，但笨拙；它知识渊博，但不会主动行动。它需要外壳，需要眼睛和手脚。

这让我想起国内一众AI助手的窘境。它们大多还停留在“更聪明的Siri”阶段，比拼的是谁的回答更像人、谁的响应速度更快。但用户很快就会发现，这些花哨的对话无法解决任何复杂问题。你让AI助手订一张机票，它能给你一段文字描述，但最终你还是得自己打开应用去操作。ChatGPT的这次转向，等于是在说：别光聊天了，得干活。

然而，OpenAI的路径是否就正确？把所有功能塞进一个应用，会不会造成另一个“超级臃肿软件”？历史上，从微软的Office到各种国产“全家桶”，这种试图一站式解决所有需求的野心，往往以复杂、臃肿和创新停滞告终。AI的灵活性在于其模块化，一个强大的核心模型，完全可以驱动无数个轻量级、专业化的应用。强求大一统，可能违背了技术分化的自然趋势。

更值得玩味的是小米的消息。当ChatGPT在奋力“外扩”时，小米的自研大模型和AI助手是否也面临同样的抉择？是继续深耕手机、智能家居这些核心场景的深度整合，做一个“无处不在的环境智能”，还是试图做一个通用的、无所不能的超级助手？两者路径截然不同。小米的优势恰恰在于其硬件生态，这或许是比做一个通用聊天窗口更现实的护城河。

说到底，ChatGPT的改版是一个风向标，标志着AI竞争从“模型擂台赛”进入了“产品落地攻坚期”。谁能把AI能力无缝地、无感地编织进用户的具体任务流里，而不是强迫用户学习新的交互方式，谁才可能赢得下一轮。从这个意义上说，小米们的机会或许更大，因为它们离用户的“具体生活”更近。而OpenAI，则面临着从“技术神坛”走向“应用红海”的惊险一跃。这场改造，成则封神，败则可能成为又一个被自己创造物所困的巨人。