After Meituan's Tabbit international version launched, it opened multiple advanced large models including GPT-5.5 and Opus 4.8, along with basic agent functions, for free use. Most such models previously adopted a paid subscription model with no small monthly cost, so the free benefit has clear appeal to high-frequency users and naturally becomes the product's user-acquisition hook. This looks a lot like the familiar subsidy-driven user acquisition of the internet era: trading high-value benefits for user scale and winning the market through scale. But if you take the business logic apart further, you find that the free model of AI products and the traffic playbook of the internet era are not the same thing at all—from the underlying mode of production to the logic of competition.
The Foundations of Two Kinds of Free: An Essential Difference in Mode of Production
The core premise behind the internet's free model is that the marginal cost of replicating a digital product approaches zero. Once a codebase and service architecture are complete, adding each new user incurs almost no extra production cost. As the user base grows from one hundred thousand to ten million, the increase in marginal cost is extremely limited. On this basis arose the complete closed loop of "free in exchange for traffic, traffic converted into revenue": using free features to attract user attention, then realizing commercial monetization through advertising, value-added services, and the like. Throughout this logic, the user's core identity is traffic—a carrier of attention—and scale is the core competitiveness. The production logic of AI products is entirely different. Every conversation generated, every agent task executed, requires calling on compute to complete real-time inference, corresponding to real Token costs and compute consumption. The more users there are, the higher the usage frequency, and the more complex each user's tasks, the more the vendor's total spending rises in tandem; there is no possibility that "scaling up dilutes the core cost." This means that relying purely on free user acquisition cannot directly form a commercial closed loop in the AI race. If you simply copy the internet's traffic playbook, the larger the user base, the worse the losses become.
From Grabbing Traffic to Grabbing Accumulation: Redefining User Value
The difference in mode of production ultimately points to a change in how user value is defined. In internet logic, the core asset users leave behind is shallow behavioral data such as clicks and views, and value concentrates on the secondary monetization of attention. Users are fluid—using one product today and switching to another tomorrow—and traffic itself has no compound effect; the product must continuously rely on new features and new content to retain user attention. In AI logic, each of the user's questions, each correction of an output, and each adjustment of a task flow is behavioral data from a real scenario. This data directly feeds back into the product: optimizing task-execution accuracy, adapting to the user's habits, and fitting the user's decision preferences. The product experience keeps improving with use, and the user data itself has a compound effect. Correspondingly, the retention logic changes as well. Internet products have very low switching costs; changing tools mostly costs shallow usage habits. The switching cost of AI products, by contrast, rises gradually with time spent using them: the longer one uses it, the more personal habits, task flows, and historical data accumulate in the product, and switching to a new product means restarting a great deal of personalized accumulation from scratch. This also explains why AI vendors are willing to bear high compute costs to offer free tiers: what they want in exchange is not one-off traffic, but user data that keeps generating value, plus the workflows bound to the product. Traffic passes through; it is the accumulation of workflows and data that forms the real competitive moat.
An Industry Consensus: Free Is the Entry Ticket, Accumulation Is the Endgame
Tabbit's choice is not an isolated case. Broadening the view, multiple tracks—office collaboration, general large models, and more—have almost all adopted the strategy of opening basic capabilities for free. What looks like separate choices across different tracks actually shares a highly consistent underlying goal: using free access to lower the barrier and seize users' workflow accumulation. In the AI-browser track, the core aim of opening top-tier models for free is not to seize the market share of traditional browsers, but to guide users to migrate workflows such as web information aggregation, cross-site operations, and research into the AI browser. Only when users' task-operation habits and scenario preferences accumulate in the product does genuine stickiness form. In the office-collaboration track, products such as DingTalk Wukong, Feishu AI Companion, and WPS AI all open basic AI capabilities for free to individual users and small and medium-sized enterprises. This is essentially lowering the access barrier to AI office work; once a company deeply binds core office scenarios—meeting minutes, document processing, report analysis, workflow approval—to the platform's AI capabilities and deposits the team's collaboration data and knowledge system on the platform, the subsequent replacement cost rises substantially. In the general large-model track, vendors commonly launch permanently free basic versions covering lightweight scenarios such as daily Q&A and simple creation. On one hand this permeates users' daily habits, letting heavy demand flow naturally to paid tiers; on the other, massive volumes of interaction data continuously feed back into model-capability iteration, forming a positive loop.
Conclusion
"Paid lobster cools, free lobster rises"—on the surface it looks like another round of industry price war, but in reality it is a quiet switch in the AI industry's rules of competition. The free of the internet era ended at traffic scale and attention monetization, with the core of competition being the fight for the entry point; the free of the AI era merely lowers the barrier for users to enter, and the true competitive focus is the accumulation of users' workflows and data assets. When traffic is no longer the core yardstick, the depth of workflow accumulation becomes the real moat of the next competitive phase.
