A series of complex instructions spanning office, education, and information management scenarios, such as "I need to generate a research report on China's embodied AI industry for 2026...", "Help me create a middle school biology PowerPoint on 'Photosynthesis'...", and "Push targeted financial news from the past 24 hours...", are all tasks that Doubao Pro aims to handle in a "one-stop" manner.

Doubao Pro officially launched on June 24, adopting a three-tier pricing structure: Standard at 68 yuan/month, Enhanced at 200 yuan/month, and Premium at 500 yuan/month. As a leading product with a massive monthly active user base of 345 million, the largest among current domestic consumer-grade large language models, this move by Doubao presents a real-world question to hundreds of millions of ordinary users: As AI transitions from a "toy" to a "tool", how much real money are you willing to pay for it?

According to official information, Doubao's Office Task Mode integrates the Doubao 2.1 series models capable of executing Agent tasks. This mode supports operations on local computers, using a browser, calling Skills, and scheduled tasks. It comes with a built-in Office suite and supports professional image and video design and generation, as well as sharing application websites. Meanwhile, free users can also experience the Office Task Mode powered by the Doubao 2.1 Turbo model.

To this end, tests were conducted on Doubao's Office Task Mode and Pro features, covering multiple scenarios including local file management, industry report generation, PowerPoint creation, data visualization, and scheduled push notifications. The experience differences between the free version and the Standard paid tier were also compared.

Opening Doubao's subscription page reveals that the core differences between the three tiers are concentrated in the quota allocations for Office Tasks, Expert Mode, and image/video generation functions.

At the model level, paid users can use the latest Doubao 2.1 Pro model in Office Task Mode, while free users are limited to Doubao 2.1 Turbo. Additionally, paid subscriptions come with benefits like larger context windows and more generous quotas for scheduled tasks.

Do higher model permissions and larger quotas translate to a substantive improvement in office efficiency? With this question in mind, testing began after subscribing to the Standard tier.

"Local file operations" are a foundational capability of Doubao's Office Mode. In the first test, it was instructed to move images from a MacBook's download directory to a specified folder and perform secondary classification based on the month each image was saved. The task took approximately 3 minutes, during which two manual authorization requests were triggered due to macOS permission mechanisms. Ultimately, all 327 images in the folder were successfully transferred and archived under a unified "Year-Month" naming format.

The second test focused on industry content generation. An instruction was given to Doubao in Office Task Mode to write a "2026 China Embodied AI Industry Research Report", explicitly requiring coverage of four sections: market size, key players, technology roadmaps, and application scenarios. During the test, Doubao performed multi-source data searches and spent about 3 minutes generating a well-structured document exceeding 5,000 words with clear data presentation. When asked to supplement with company introductions, Doubao was able to flexibly add relevant material to the existing content.

To further verify the data accuracy of this report, five or six data points were randomly sampled and manually verified; all were found to be accurate. However, some industry data cited older forecast values, resulting in suboptimal timeliness and somewhat diminishing the report's reference value. Reports were also generated from both free and paid user accounts for comparison. The overall completion quality showed little difference, with the free version even presenting the key players in the sector in a table format.

It is worth noting that Doubao Pro, while generating the online document, also provided a Feishu document link. Users can click the blue text to enter Feishu for online editing, and modifications will sync automatically. The free version, although it completed the report writing, did not offer this Feishu integration feature.

The third test evaluated PowerPoint creation skills. An instruction was given to create a middle school biology PowerPoint on "Photosynthesis", requiring it to include four parts: principle explanation, classic experiments, influencing factors, and life applications, with images on each page. From the presentation results, both the free and Pro versions basically met the framework requirements for the teaching material and allowed for slide show and editing within the Doubao interface, but each had its own shortcomings in detail.

Issues with the free version's PowerPoint centered on content adaptability; knowledge point descriptions were somewhat obscure for the seventh-grade level, lacking extensibility and interactivity, and the teaching notes in the comment section were rather rigid. The Pro version's PowerPoint showed clear optimization in logic and expression, sounding more like classroom instruction. However, regarding image support, the entire Pro presentation lacked any real teaching illustrations or actual scene photos. In contrast, the free version not only made extensive use of standard schematic diagrams from textbooks and academic literature but also paired real-life photos in the life application section, making it more practical for teaching.

However, the Pro version's "image shortcoming" showed significant improvement in a second follow-up test. When the same prompt was used again to generate a PowerPoint, the Pro version not only supplemented teaching illustrations and real-life photos but also embedded dynamic graphics in the life application section.

The fourth test focused on data visualization tasks. Screenshots detailing the features and quotas of each Doubao Pro tier were organized into a folder, and Doubao was instructed to extract the image information and compile it into a comparison chart. The initial output contained several discrepancies; some information was not accurately identified, and even several exclusive services not present in the screenshots were added.

However, Doubao's error correction capability performed reasonably well. After being asked to verify each item against the screenshots one by one, Doubao sequentially retrieved the corresponding images for re-verification. The corrected content matched the original information in the screenshots.

The fifth test centered on scheduled content push functionality. Doubao was instructed to generate a financial industry daily report at a specified time, summarizing financial news published by authoritative media in the past 24 hours and annotating the information sources and original links. Ultimately, Doubao pushed the requested daily report content precisely at the scheduled time, demonstrating stable execution of the scheduled task.

Overall, Doubao performed commendably in core office tasks like file organization and report writing, but there is still room for improvement in areas like visual recognition and PowerPoint image selection.

Given that the free basic version can meet daily needs, will users be willing to pay extra for stronger execution and deeper integration? The answer to this question ultimately depends on whether the product's capabilities can match its pricing ambitions.

Doubao is not the first product in China to test paid services for the consumer end. Baidu's Ernie Bot and MoonShot AI's Kimi have previously launched individual subscription services. However, the domestic market's base pricing is generally below 50 yuan per month; for example, the entry tiers for Kimi and Zhipu AI's ChatGLM are both 49 yuan/month. The main price range of 80 to 200 yuan per month gathers the advanced versions from most providers.

Leading overseas products started their commercialization paths earlier and have clearer tiering: entry-level subscriptions like ChatGPT Go are priced at $8/month; mid-tier options like Google's Gemini Advanced cost $19.99/month; OpenAI's ChatGPT Plus is also $20/month. There are also heavy-usage tiers ranging from $100 to $200, with clear differentiation in quotas and features.

While the large language model wave sweeps across various industries in China, the path to AI commercialization remains in an ongoing exploration phase. The direct driving force accelerating this process is the dual pressure of costs and growth borne by the entire industry.

According to market information, ByteDance's AI infrastructure expenditure by 2026 could reach 200 billion yuan. As of the time of writing, ByteDance has not officially responded to this. However, looking across the industry, high investment and significant losses have become the norm.

Zhipu AI's (HKEX: 02513) 2025 financial report shows annual total operating revenue of 724 million yuan, a year-on-year increase of 131.9%; R&D expenses were as high as 3.18 billion yuan, with an annual loss of 4.718 billion yuan, expanding by 59.5% year-on-year. The core reason for the expanded loss was the continued increase in strategic R&D investment. Meanwhile, in the first quarter of this year, Zhipu AI's API call pricing also saw a significant increase.

Against this backdrop, the official launch of Doubao's paid version has sparked widespread market attention and discussion.

Cui Lili, Vice President of the Digital Economy Research Institute at Shanghai University of Finance and Economics, stated in an interview that the current appeal of Doubao Pro is still insufficient. The current market is rich with various conversational AI apps and Agents, and users find it relatively convenient to call APIs for self-deployment. The truly heavy-paying user groups also typically adopt this self-built API calling method. In her view, in the video field where ByteDance excels, choosing to pay for Doubao is not as good as opting for more vertical products like Jianying (CapCut). The significance of Doubao Pro lies more in reducing token waste, cultivating payment habits, and providing feedback for product refinement.

Pan Helin, a well-known economist and member of the Expert Committee on Information and Communication Economy of the Ministry of Industry and Information Technology, also noted in an interview that Doubao's pricing is not problematic, but there are indeed many similar office products currently available.

"Previously, Doubao + Jianying + Jianying (CapCut) were tailored for content creators. This expansion into the office domain indicates that Doubao wants to develop more new users in the B2B sector to explore new AI customer acquisition channels," Pan Helin believes. He considers Doubao Pro's pricing reasonable because "B2B users are actually not very price-sensitive. However, since there are many alternative products, the reason for B2B users to accept Doubao is a question that requires deep thought."

As industry competition shifts from a race for user scale to a comprehensive contest of scenario value and commercialization capability, major providers need to reduce unit costs through inference optimization and computing power scheduling on one hand, and on the other hand, build competitive moats in vertical scenarios like office work, content creation, and coding.

Moreover, as users begin to pay for AI, their demands for the accuracy and professionalism of the deliverables also rise to a new level. This explains why market discussion remained heated after Doubao's paid service announcement. On one hand, Doubao's leading user activity indicates its convenience for daily use has gained broad recognition. On the other hand, its professionalism and accuracy have not yet formed sufficient consensus in the market.

For users, the pricing of a paid version should not merely reflect an upgrade in model capability but must also provide accurate, reliable, and usable feedback at the product experience level. From the current perspective, although Doubao Pro can deliver products with good performance, it still requires manual verification when facing complex business scenarios. There is still a gap before users can completely become "hands-off managers." In light of this, whether users are willing to pay for it remains an open question.

For Doubao, relying on ByteDance's computing power investment and ecosystem matrix, tiered commercialization has taken its first step. However, the transformation from a traffic product to a productivity tool still has a long way to go. How to convert comprehensive functionality into sustained willingness to pay, and how to bridge the conversion path from consumer to business users, will be the core questions it needs to answer next.