Doubao has officially rolled out its paid version today, featuring three distinct pricing tiers:

On June 23, Volcano Engine unveiled its latest Doubao 2.1 series large models at the FORCE conference, boosting capabilities in coding and agent tasks.

When questioned about Doubao's monetization strategy, Tan Dai, President of Volcano Engine, offered a cryptic reply: "Doubao isn't a core part of Volcano's business, but as far as I know, it will continue to offer free, high-quality services to users. Simultaneously, it will soon introduce a professional office task mode tailored for productivity scenarios, powered by our newly launched 2.1 Pro model."

A deeper analysis of Tan Dai's statement and Doubao's official announcement reveals that the emphasis lies not on the specific pricing details but on reassuring users that Doubao will maintain its free service offering to a broad audience, with free users also gaining access to new model capabilities.

Prior to the official announcement of the paid policy, media reports suggested that Doubao would introduce a paid version in late June, sparking immediate debate. The resistance may stem from the difficulty in accepting that the quirky and endearing Doubao has evolved into another 'token-consuming' service.

However, Doubao has reached a juncture where charging is inevitable. According to Volcano Engine, as of June 2026, Doubao's large models process over 180 trillion tokens daily, marking a more than tenfold increase over the past year. Without revenue, these usage levels would incur substantial costs that ByteDance cannot overlook.

Now that monetization is unavoidable, the primary challenge lies in determining how to price the service without alienating Doubao's hard-earned user base.

I. Pricing: Just a Small Hurdle

Doubao's approach to launching its professional version has been relatively cautious, considering the psychological impact on its vast free user base. However, to some extent, the free model that fueled Doubao's growth has also constrained its path to commercialization.

Currently, the paid version of Doubao faces several hurdles.

In terms of pricing, Doubao is positioned among the highest in the AI subscription market. The basic tier is priced at 68 RMB/month, offering over five times the usage limits of the free version for features like expert mode and office task mode. The enhanced package costs 200 RMB/month, providing four times the limits of the standard package, while the premium package costs 500 RMB/month, offering ten times the limits of the standard package.

Annual subscriptions are priced at 688 RMB, 2048 RMB, and 5088 RMB, respectively.

In comparison, Kimi offers multiple tiers starting at 49 RMB/month; Zhipu Qingyan's VIP monthly subscription is 79 RMB, with its SVIP priced at 229 RMB; Minimax's lowest tier is also 49 RMB/month. Doubao's basic tier is nearly 20 RMB higher than Kimi's entry-level tier, and its 500 RMB premium tier competes directly with the highest-priced productivity tools. Meanwhile, Alibaba's Qianwen remains free, with its flagship model Qwen3.7-Max available at no cost across all scenarios. Tencent Yuanbao has also stated it has no plans to charge, focusing its commercialization efforts on the Hunyuan API and enterprise services, using corporate revenue to subsidize free personal use.

The decision-making process for office-related expenses is straightforward: users will pay if the model consistently delivers satisfactory results.

However, it's undeniable that the performance of general tasks across various products has not yet shown significant generational differences. Even if Doubao outperforms competitors in certain tasks, whether ordinary white-collar workers will pay nearly 20 RMB more per month for a 10% or 20% performance boost remains uncertain.

Of course, coding and agent tasks represent a proven, rapidly commercializable market. A research report by CMB International in April this year stated that Anthropic's Claude Code generated $1 billion in annual recurring revenue (ARR) within six months of launch, increasing to $2.5 billion by February 2026.

In this field, Doubao is a latecomer. Competitors have already established user loyalty and provided tangible benefits. Doubao not only needs to convince users that it offers not just emotional value but also office productivity value but also faces the additional challenge of shedding its humorous and abstract image while positioning itself as professional and efficient. In the serious realm of coding and agent tasks, Doubao's reputation for being fun, witty, and even a bit clumsy feels out of place.

These are inevitable transformation costs that Doubao's professional version must bear in acquiring users, placing even higher demands on the technical prowess of Doubao's models.

Notably, after the launch of Doubao's professional version, its positioning within ByteDance has become somewhat ambiguous.

Previously, ByteDance's AI business units had clear roles: Doubao served as the entry point, Kouzi focused on agents and workflows, TRAE provided programming tools, Feishu targeted enterprise collaboration scenarios, and Seed offered foundational models and multimodal capabilities.

The introduction of Doubao's professional version essentially consolidates capabilities previously dispersed across various business units into Doubao, the mainstream entry point. It not only fills gaps in coding capabilities but also highlights long-chain agents and office expert modes as core selling points. Ordinary users no longer need to switch to Kouzi or TRAE; they can complete the entire process—from requirement understanding and workflow construction to code implementation—within Doubao.

This naturally raises new questions about internal positioning: Now that Doubao directly incorporates agent construction capabilities previously core to Kouzi, will Kouzi's role shift to becoming the foundational capability provider behind Doubao? Similarly, with TRAE's programming capabilities now supported by Doubao's professional version, where do the user boundaries lie between the two? Furthermore, with Feishu primarily focusing on office AI scenarios and Doubao's professional version also emphasizing productivity tools, will the two products end up competing internally?

From this perspective, the public controversy sparked by pricing may be a relatively minor obstacle on Doubao's path to commercialization. As Doubao transitions from being just a consumer toy for chatting with users to a product with greater strategic significance, it is bound to face more complex and far-reaching challenges.

One such challenge is that if Doubao becomes the external window for all of ByteDance AI's capabilities, it will inevitably create overlaps or even internal conflicts with the original vertically oriented products, making the challenge of balance an ongoing issue.

II. Doubao Steps Out of Its Comfort Zone

Doubao has always had a somewhat paradoxical character. It seems to exist on the fringes of mainstream AI applications, resembling more a classic internet product powered by AI.

Since its inception, Doubao has followed a free model, providing rare emotional value to hundreds of millions of users through its charming yet slightly clumsy persona. This approach has earned it massive traffic and high user engagement. As of the first quarter of 2026, Doubao had approximately 140 million daily active users and 345 million monthly active users, surpassing the combined total of the second and third-ranked apps.

This impressive performance undoubtedly grew out of ByteDance's familiar product strategy. Doubao inherited the user acquisition methods used by Douyin and Toutiao, becoming a product nurtured by ByteDance's most effective traffic-driving tactics and supported by company-wide resources.

Doubao has also served as ByteDance AI's public face. Its only rebranding occurred before its public beta test. In March 2023, after ByteDance established the Flow department, Zhu Jun led the development of Doubao under the internal codename 'Grace.' However, six months later, the less personable 'Grace' was renamed 'Doubao' before its official public test.

Zhu Jun once revealed the three principles behind naming it Doubao: personification, closeness to users, and individuality. To maintain Doubao's down-to-earth and lively feel, ByteDance even deliberately adjusted the product's expression logic, making Doubao's responses seem not overly intelligent but sufficiently sensible. Even though ByteDance underwent large-scale organizational restructuring due to falling behind competitors in AI development, Doubao's approachable public image remained unchanged.

These achievements, driven by traffic-focused strategies, have now created a backlash as Doubao prepares to charge. The challenge of shifting user expectations from free to paid is both inevitable and difficult: there may never be a 'perfect' time to convince users to pay.

This is partly because Doubao has not yet found a suitable way to monetize its traffic. Previously, Doubao's AI-powered e-commerce remained superficial, with gray-test in-conversation shopping features that, while integrated with Douyin's e-commerce ecosystem, never scaled significantly. Attempts to expand into local services and ride-hailing also failed to gain traction.

Since these traditional internet traffic monetization models did not pan out, Doubao ultimately chose the most certain—and hardest—path: professional office scenarios as its commercialization breakthrough.

Until today, Doubao can be said to have truly stepped out of the comfort zone created by ByteDance, preparing to compete head-on with rivals in the productivity arena. Liang Rubo explicitly conveyed ByteDance's commitment to AI at yesterday's FORCE conference via video: scaling AI peaks is the company's top priority. Doubao has become a crucial pillar in this endeavor.

In the past, ByteDance used its most effective traffic-driving strategies and organizational resources to elevate Doubao into China's most widely used AI application, giving it the highest starting point and best resources, making it the company's 'darling.'

However, ByteDance's patience with Doubao may not last much longer. According to media reports, over the past six months, ByteDance has begun scaling back AI businesses that have not yet shown commercial returns. In February, the company internally halted the Doubao AI glasses project due to difficulties in creating truly differentiated capabilities. In June, Lin Xi, head of Doubao's mobile hardware products, left the company, followed by Gu Quanquan, a core member of ByteDance's Seed pre-training team, who also announced his departure. His AI4S team underwent another round of organizational restructuring.

This is closely tied to soaring computational costs. As of May 2026, Doubao's daily computational expenses reportedly reached tens of millions of RMB. ByteDance's AI capital expenditure plan for 2026 has been raised to over 200 billion RMB, roughly equivalent to 60% of the company's total profits in 2025. However, daily platform revenue during the same period was less than one million RMB, with nearly all income coming from e-commerce commissions.

To cover these escalating costs, ByteDance will inevitably withdraw its full support for Doubao. In today's AI landscape, even ByteDance may struggle to sustain the massive, prolonged consumption caused by AI, and its patience for unprofitable businesses will not last indefinitely.

Given the group's expectations, a Doubao that can quickly achieve a self-sustaining commercial closed loop may be even more 'adorable.'