Woofun AI reports that Zhipu officially launched three paid subscription tiers on June 24, pricing the Standard Edition at 68 yuan per month, the Enhanced Edition at 200 yuan per month, and the Premium Edition at 500 yuan per month, all powered by the Zhipu 2.1 series models. Simultaneously, Baidu announced the consolidation of its Wenxin Yiyan Web, Wenxin, and Wenxin Assistant platforms into a single Wenxin Assistant interface while maintaining a strictly free pricing model. These divergent moves by two industry leaders highlight a fundamental fracture in the prevailing logic of the mobile internet era, where expanding entry points was synonymous with capturing traffic and securing bargaining power. The core question facing the sector is whether this historical rule remains valid when the cost structure of AI products fundamentally differs from traditional internet applications, where every interaction consumes tangible hash rate resources rather than negligible server bandwidth. Zhipu and Wenxin represent the two extreme poles of this new economic reality, with one monetizing specific high-compute tasks and the other expanding capabilities as free infrastructure to lock in user workflows. The near-simultaneous convergence of these strategies offers a critical lens through which to observe the commercialization trajectory of domestic large-scale AI models.

The proliferation of excessive entry points has emerged as a defining 'midlife problem' for nearly every AI product company, stemming from an early expansion logic that prioritized securing market position before refining user experience. This approach, while effective for initial user acquisition, has resulted in a fragmented landscape where users struggle to locate the appropriate tool for their needs, a situation vividly illustrated by Microsoft's Copilot ecosystem. Users currently face a disjointed experience where Copilot exists as a separate entity within Word, Teams, GitHub, and the Windows operating system, requiring navigation across five distinct entry points that fail to share conversation history or context. This redundancy forces users to manage their own memory across disconnected interfaces, turning what should be an assistive tool into a cognitive burden. While such fragmentation was tolerable during the early chat-box phase where Copilot handled basic tasks like meeting summaries, email drafting, or code completion, the industry has now evolved toward intelligent agents capable of planning, executing, and managing complex tasks across multiple scenarios. Building an effective system using multiple independent assistants with identical branding is structurally impossible, as the lack of unified context prevents the AI from functioning as a cohesive agent rather than a collection of isolated tools.

As user expectations have shifted from simple conversational capabilities to the demand for actual task completion, the existence of too many entry points has become a direct obstacle to operational efficiency, prompting global giants to consolidate their scattered interfaces. Google has responded by integrating Gemini directly into the Chrome browser, initially deploying it in the sidebar and subsequently introducing 'automatic browsing' features that allow the AI to handle price comparisons, form filling, and ticket booking while retaining human control for sensitive operations like logging in or finalizing payments. The strategic objective for Google is to establish Gemini as the primary AI entry point across search, browsing, and office applications, thereby ensuring that its core search business is not cannibalized by fragmented chat-based interactions. Microsoft is pursuing a parallel strategy, with Fortune reporting in late May based on internal sources that the company is developing an unnamed 'super application' to unify Copilot Chat, GitHub Copilot, Copilot Cowork, and the internally developed Autopilot workflow engine into a single interface. Jacob Andreou, the head of the Copilot project, is leading this initiative with a target release date set for the end of the summer, aiming to eliminate the friction of users deciding which assistant to deploy before starting a task. Although the specific implementation details differ between these tech giants, the underlying problem they address is identical: the necessity of removing decision fatigue regarding AI tool selection to maximize productivity.

Baidu's decision to merge three distinct services represents a localized solution to this global challenge, where the visible reduction in complexity through domain consolidation masks a deeper architectural shift in how functions are scheduled. By closing several websites and funneling traffic to a single domain, Baidu has integrated search, document libraries, office tools, and intelligent agent capabilities into a unified scheduling system that autonomously determines which specific function to deploy rather than leaving this decision to the user. In the previous model, a user wishing to generate a report would first need to manually decide whether to query Wenxin Yiyan for ideas or search the document library for a template, a process that introduced unnecessary friction. Under the new system, the user simply states their requirements, and the backend architecture handles the routing and execution, effectively removing the cognitive load of tool selection. This transformation, while appearing as a mere user experience optimization, actually signals a profound shift in the focus of AI competition, moving from the accumulation of entry points to the ability to enable users to complete complex tasks within a single, seamless environment.

Zhipu's move to implement user charges was anticipated well before its official execution, having hinted at paid options in its App Store description a month prior to the June 24 launch. After nearly two months of market discussion, the charges were formally implemented, establishing three distinct pricing tiers: the Standard Edition at 68 yuan, the Enhanced Edition at 200 yuan, and the Premium Edition at 500 yuan, specifically targeting high-hash-rate productivity scenarios such as software development, data analysis, professional design, and long-document processing. These specific use cases were the most difficult to sustain under a purely free model due to their intensive computational requirements. When compared to global competitors, Zhipu's 68 yuan entry price is lower than the starting prices of ChatGPT Plus and Claude Pro, which hover around 100 yuan, while its 500 yuan top tier remains significantly lower than the premium pricing of ChatGPT Pro and Claude Max, which range from 200 to 300 dollars per month. This marks the first instance where domestic large-scale AI models have directly priced 'complex' functions within their consumer applications, distinguishing between simple Q&A tasks that remain free and those requiring significant hash rate and computational resources that now incur fees. This strategic choice reflects intense real-world pressures, as Woofun AI data shows that as of March this year, Zhipu's large-scale AI model was generating over 120 trillion token requests per day, a figure that doubled in just three months and reached 1,000 times the initial level recorded in May 2024.

The traditional mobile internet axiom that 'more users mean more profits' fails to apply to the AI sector, where free conversations incur direct hardware depreciation and electricity costs, meaning a larger user base can translate into heavier financial burdens rather than increased leverage. From this economic perspective, Zhipu's decision to charge users is a rational response to cost structures, even as Wenxin has chosen a diametrically opposed path driven by Baidu's long-standing commitment to 'long-termism.' Beyond the merger of services, Baidu released the Wenxin Large Model 5.1 with an unchanged pricing strategy, keeping it free for all users while expanding its feature set to include online editing of Office documents, scheduled tasks, AI-assisted report writing, AI-powered PPT creation, in-depth research, and AI music generation. The scope of Wenxin has thus expanded from basic Q&A to specific learning, office, and daily-use scenarios, a move made possible because its cost structure has been significantly reduced through technological advancements. Wenxin Large Model 5.1 achieved a score of 1,223 points in the international Search Arena evaluation, ranking fourth globally and first in China, and stands as the only domestic model on this list. In the highly competitive AIME26 evaluation, it scored 99.6 points, placing it second only to Gemini 3.1 Pro. More critically, its cost structure is highly efficient; Wenxin 5.1 utilizes a 'multi-dimensional elastic pre-training' method that extracts the optimal sub-network from the sub-model matrix of Wenxin 5.0, allowing it to inherit knowledge without starting from scratch. This optimization has reduced the total number of parameters to about one-third of the previous version and halved the number of activation parameters, resulting in pre-training costs that are only 6% of those for similarly sized models.

However, this cost advantage alone does not fully justify the concept of 'long-termism,' particularly as domestic large-scale AI models have seen a wave of charging initiatives this year, making Wenxin's decision to go against this trend a significant strategic outlier.

The broader industry landscape reveals a pattern of pricing adjustments that underscores the financial pressures facing AI developers, with Zhipu increasing the prices of its GLM Coding Plan on February 12 and eliminating the initial discount, setting the Lite edition at 49 yuan and the Max edition at 469 yuan. Inspir's Step Plan was first made available for payment on March 23, starting at 49 yuan per month, while MiniMax took a more dramatic approach on June 1 by changing its billing method from pay-per-use to pay-per-token consumption upon the launch of the new M3 model.

This shift caused the entry-level plan, which previously cost 29 yuan, to suddenly increase to 49 yuan, sparking significant controversy within the developer community. In contrast, Baidu, a company with extensive experience in setting AI prices, has chosen to keep its basic functions free, demonstrating a confidence in its long-term vision that requires not only strong cost control capabilities but also a firm belief in its strategic direction. A short-term market stance is insufficient to support such a decision, which relies on Baidu's heavy investment over the past decade in AI research and development covering all four layers of the technology stack: chips, frameworks, models, and applications. The sustainability of this free strategy ultimately depends on whether this entire stack can function effectively to generate value beyond direct user payments.

The divergent approaches taken by Zhipu and Wenxin are fundamentally determined by their respective resource constraints and strategic models. Baidu follows an 'ecosystem flywheel' model, leveraging technological advantages to attract users and building ecological barriers that can be monetized through existing services such as search and cloud computing, thereby realizing the value of AI capabilities at a distance from the user. This approach offers the advantage of generating revenue indirectly while closely integrating with Baidu's established businesses, providing strong resilience against economic cycles. Zhipu, conversely, follows a 'product flywheel' model where product capabilities generate direct user payments that fund further model development, creating a faster cycle of growth that is more user-centric but requires continuous proof of value. The success of Zhipu's model is measured directly by whether users choose to renew their subscriptions monthly, whereas Baidu's model relies on the broader ecosystem to capture value. In terms of user strategy, Wenxin aims to attract a massive number of daily active users by offering a wide range of functions, adhering to a 'wide coverage' approach, while Zhipu adopts a tiered operation model using the free version to attract users and the paid version to generate revenue, following a 'targeted user' approach. Both strategies aim to match users with the value they provide but do so through different mechanisms: one focuses on expanding the user base first and then selecting valuable functions, while the other sets a price for specific functions and lets users choose according to their needs.

Furthermore, the ways these two approaches convey technical confidence differ significantly; Wenxin offers top-tier functions for free to reflect its engineering capability to continuously improve models, while Zhipu charges for advanced functions to define their value through pricing. Both utilize 'free' and 'paid' as entry points to build user trust, with the ultimate goal of convincing users that the company's technical capabilities are worth the investment.

Although the starting points and scales of these two approaches differ, they are not necessarily mutually exclusive, and both choices make practical sense in the short term given their respective resource structures. Baidu possesses an ecological foundation that supports its free model, while Zhipu needs to quickly establish a commercial model to sustain its operations. Neither approach is inherently superior; rather, they reflect different resource structures and development strategies tailored to each company's specific position in the market. In the medium to long term, the real determinants of success will hinge on three evolving factors: how much further hash rate costs can be reduced, to what extent users are willing to pay for AI services, and whether AI can truly become an irreplaceable productivity tool. All three of these factors remain in flux, making it premature to draw definitive conclusions about which model will prevail. Globally, these two approaches are becoming a common philosophy in product development, with even companies like Google and Microsoft, which have consolidated their entry points, leaving room for flexibility in pricing. Microsoft 365 Copilot offers a monthly subscription of 30 dollars for enterprise customers while maintaining free basic functions for consumers, and Gemini offers multiple pricing tiers, including a reduction of its top-tier Ultra subscription from 249.99 dollars to 199.99 dollars per month and the addition of a 99.99-dollar intermediate tier following its I/O conference in May. The free version remains available in Google's product portfolio, reinforcing the consensus that consolidating entry points first and then adjusting pricing based on scenarios and user groups is the emerging standard. From this perspective, free and paid models are not mutually exclusive but represent different approaches that a company may need to adopt at different stages of development and for different user segments.

The simultaneous differentiation between Zhipu's tiered pricing and Wenxin's free expansion highlights a critical realization: entry points are becoming a scarce resource in the AI industry. Over the past two years, the industry operated under the assumption that entry points were unlimited, leading companies to launch multiple websites for a single product or embed multiple functions in different assistants with the expectation that users would navigate the complexity.

However, daily usage habits are highly ingrained, with most people allocating a limited amount of time to only one or two AI applications. Once a particular entry point becomes the preferred option for completing a task, other similar products may never get a chance to be used, regardless of their technical merits.

Moreover, the cost of switching between AI applications is significantly higher than anticipated, as user dependence stems from usage habits, historical data, file storage, task context, and trust in the specific tool. The higher the switching cost, the greater the value of securing that entry point in advance. Therefore, when Wenxin combines three entry points into one and extends its functions to office and learning scenarios, it is essentially using lower barriers and a smoother experience to gain a foothold and gradually establish a larger commercial foundation. In the end, the competition in AI will not necessarily depend on who can convert users into paying customers first, but rather on who can serve more needs at a lower cost and who can establish a lasting competitive advantage in users' workflows and daily routines. Consolidating entry points is merely the beginning, and tiered pricing is just one aspect of commercialization; the real battle is yet to begin. This marks a definitive shift where the metric of success is no longer traffic volume but the depth of integration into the user's productive life.