Doubao Pro Launches with Paid Subscription: Agent Office Task Testing and Pricing Analysis

On June 24, 2026, ByteDance's AI app Doubao officially launched its Pro version, ending the all-free era, and simultaneously released the Doubao 2.1 series of large models. The Pro version offers three subscription tiers: Standard at 68 yuan/month, Enhanced at 200 yuan/month, and Premium at 500 yuan/month, with a student discount of 38 yuan/month. The core change is the addition of an 'Office Task Mode,' powered by the Doubao 2.1 Pro model, which enables the Agent to autonomously break down goals, invoke local computer, browser, Feishu Office Suite, and other tools to deliver finished products (e.g., reports, PPTs, websites). Free users retain existing features, with the underlying model upgraded to Doubao 2.1 Turbo, and can have limited access to the Office Task Mode.

Pricing and Market Positioning

Three subscription tiers: Standard 68 yuan/month (5 hours of task quota), Enhanced 200 yuan/month (20 hours), Premium 500 yuan/month (50 hours).
Comparison with overseas: GPT Plus $20/month (approx. 140 yuan), Claude Pro $20/month; Doubao Standard is cheaper, and the high-end tier is also lower than the overseas top tier ($100-200).
ByteDance stated that free users' daily usage will not be affected, and the Office Task Mode will only offer more quota to paid users.

Office Task Mode Performance Tests

Multiple media outlets conducted various scenario tests on Doubao Pro, covering information gathering, PPT generation, website building, financial report analysis, and literature review.

Scenario 1: Automatic Industry Weekly Report Generation

After inputting instructions, Doubao automatically retrieved internet information and generated a well-formatted Feishu document, including a summary of weekly highlights and major product updates from big tech companies, with key data and cited sources.
Tests showed that Doubao accurately captured recent major events (e.g., DeepSeek financing, GPT-5.5 release), with clear time and source annotations.

Scenario 2: PPT Creation

Given a dataset containing MAU, DAU, user feedback, etc., Doubao generated a structurally complete PPT in about 10 minutes, extracting a main theme of 'AI features driving growth, but experience issues affecting retention,' including core conclusions, data cards, and problem attribution.
Visual design was acceptable, but there were issues like duplicate last two pages and some inferred data, requiring manual review.
Premium users consumed only 3% of their quota for one PPT.

Scenario 3: Website and App Generation

Provided with a reference website (e.g., booooooom.com), Doubao generated a responsive personal portfolio site including homepage, works showcase, and about me, and automatically deployed it.
In another test, based on a description of an 'AI blind box travel machine,' Doubao directly produced a complete demo with itinerary, budget, and attraction recommendations, but this exhausted the Standard tier's 5-hour quota.

Scenario 4: Financial Report Analysis and Investment Summary

Doubao extracted data from PDF financial reports, such as Nvidia's Q1 2026 revenue of $81.6 billion (up 85% YoY) and AMD's revenue of $10.3 billion (up 38% YoY), generating a comparison table and a 500-word investment summary with accurate data and industry trend insights.
The document was marked 'Partially generated by Doubao,' emphasizing the need for manual verification.

Scenario 5: Literature Review

Doubao searched for papers from ByteDance's Seed team on arxiv, OpenReview, etc., automatically downloaded and read them, and after 40 minutes generated an 8,000-word review categorized by topic (foundation models, embodied intelligence, voice generation, etc.), summarizing the team's technical characteristics.

Scenario 6: Local Computer Operations

With authorization, Doubao could operate local files, e.g., writing a Python script to detect duplicate photos and retain the best quality ones. During the process, it autonomously fixed dependency issues and algorithm bugs, ultimately delivering a command-line version, GUI version, startup script, and usage instructions.

Model Capabilities and Limitations

Doubao 2.1 Pro demonstrated strong instruction following and tool invocation in complex tasks, but exhibited an 'overthinking' tendency: repeatedly iterating on a detail, causing the Enhanced tier's 5-hour quota to be exhausted.
In some tests, Doubao automatically published content without user confirmation of the final output, indicating a need for improved user authorization mechanisms.
In a programming test (creating an astronomical clock), the code generated by Doubao 2.1 Pro contained 5 errors and could not run, while other models (e.g., GLM-5.2, Kimi) showed similar issues, indicating room for improvement in stability under long instructions.

User Adaptation Suggestions

Free users: Sufficient for daily chat, information lookup, and simple tasks; no need to pay.
Students: The 38 yuan/month student price is suitable for course presentations, paper organization, and PPT generation.
Social media/operations: The 68 yuan Standard tier allows experiencing Agent capabilities like auto-publishing and cross-platform material organization.
Product/consulting/analysts: The 200 yuan or 500 yuan tiers are suitable for high-frequency document creation, data analysis, and industry research tasks.

The launch of Doubao Pro marks the shift of a national-level AI application from free growth to monetization. The Agent office mode lowers the barrier to professional tools, but issues like fast quota consumption and occasional overthinking still need optimization.