今天，豆包正式推出专业版

Doubao Professional Launches: Putting Agent Capabilities in 200 Million People's Pockets

Doubao has over 200 million daily active users—what does that number mean? China has roughly 1.1 billion internet users, so about 1 in every 5 internet users opens Doubao daily.

This national-level AI application launched a pretty important new feature on June 24: "Professional Version" + "Office Task Mode."

In plain language: Doubao is transforming from an "AI that answers questions" to an "AI that can actually get work done for you."

Office Task Mode: Agents Actually Entering Workflows

"Office Task Mode" is the core of this update. How is it different from the old Doubao?

Old Doubao: you ask a question, it answers. Question-answer, single-turn conversation.

New Doubao (Office Task Mode): you give it a work goal, and it autonomously breaks down steps, calls tools, executes tasks, and delivers results.

Specifically, what can it do?

• **Operate local computer**: Doubao can help you organize local files, categorize documents, and move information across applications (requires your authorization, of course)
• **Call Skills**: Doubao has a "Skills Center" where various professional capabilities can be placed (such as data analysis skills, chart generation skills, document writing skills), and the Agent automatically calls them based on task needs
• **Scheduled tasks**: you can set Doubao to automatically execute repetitive work at fixed times (say, every Monday morning at 9 AM automatically generate an AI industry weekly report)
• **Built-in Office suite**: Doubao can directly generate Word, Excel, and PPT files
• **Generate online applications with backend databases**: this is the most impressive—you describe a requirement, and Doubao helps you generate a website or Web application that can actually be deployed, complete with a database

A 200M-DAU App Starts Actually Working

The most noteworthy aspect of this update isn't "how impressive Doubao's technology is," but rather "Doubao put Agent capabilities into an App that 200 million people open daily."

Previously, to use Agent tools, you had to use specialized products like Claude Code, Codex, or Manus—their users are most likely programmers, product managers, and tech enthusiasts.

Now, an administrative staff member who's never written code, an undergraduate student, or a small business owner who isn't tech-savvy can access Agent capabilities just by opening Doubao.

This expansion of audience coverage might be the most important milestone in AI普及 (democratization).

Free Users Aren't Affected Either

Doubao specifically emphasized: free users' existing features and quotas are not affected, and the underlying model has been upgraded to Doubao 2.1 Turbo, with capabilities actually improved.

Office Task Mode is also available to all users with a certain quota for free. Free users can experience Office Task Mode with the Doubao 2.1 Turbo model.

In other words, this update isn't "locking good features behind a paywall"—it's "on top of the free version, providing professional users with stronger compute power and higher quotas."

Pricing: Starting at ¥68/Month, ¥38/Month After Student Verification

Doubao Professional has three tiers:

• **Standard Plan**: ¥68/month (auto-renewal), ¥38/month after college student verification, for 6 months
• **Enhanced Plan**: ¥200/month
• **Premium Plan**: ¥500/month

The difference is mainly in "usage quota"—the Standard plan has a monthly limit on Office Task invocation counts; Enhanced and Premium have higher quotas. Features and models are identical across all tiers.

The ¥38/month student price is, honestly, quite sincere. A college student spending ¥38 can access an office Agent equipped with a flagship LLM—that price is cheaper than most problem-set App memberships.

Doubao 2.1 Pro: What Level Has It Reached?

The Professional version is equipped with Doubao's latest 2.1 Pro model. According to data published by Doubao, 2.1 Pro's capabilities can already be benchmarked against Claude Opus 4.6.

A few specific metrics:

• In developer crowdsourced evaluations targeting real Coding scenarios, Seed 2.1 Pro achieved a 59.1% win rate against Claude Opus 4.6
• In the ALE (Agents' Last Exam) benchmark, Seed 2.1 Pro is in the top tier
• In the GDPval benchmark (measuring model completion quality and economic value in real work tasks), Seed 2.1 Pro scored the highest among evaluated models
• Terminal Bench 2.1 score: 71.0, essentially tied with Claude Opus 4.7's 71.7

There's also a very telling Demo: a chip design RTL test where Doubao 2.1 Pro ran continuously for nearly 18 hours, completing 6 core modules and 1,303 lines of RTL code, going through 9 rounds of iteration, and running through the complete engineering flow including simulation, testing, and synthesis checks. Previously, completing this task required 3-5 engineers working for weeks.

The Flywheel Effect of 200 Million DAUs

Doubao's 200 million DAUs bring more than just "pride"—there's a very tangible thing: the data flywheel.

200 million users processing real work tasks daily in Office Task Mode generate massive interaction data that continuously feeds back into the model's Agent capability iteration. User scale brings data scale, data scale accelerates model evolution, and model evolution further drives more users to adopt Agent features.

Doubao says the Seed Evolving series (rapidly iterating versions based on 2.1) updates every one to two weeks—this iteration speed cannot be achieved without the support of massive user data.

New Features Coming Soon

Doubao also previewed some features coming soon:

• Skills store page (supporting users adding/managing Skills)
• Remote mobile control of computer
• GUI operation support on desktop (meaning the Agent can directly operate computer interfaces, not just call APIs)
• Further integration with Feishu

If all these features launch as scheduled, Doubao's Agent capabilities will evolve from "can get work done" to "can get many kinds of work done."

What This Means for the Entire Industry

I think the real value of Doubao's update lies in: it proves that Agent capabilities can be "democratized"—not only people who understand technology can use them; ordinary users can also get started with zero barrier.

If a 200M-DAU App can truly make Agent workflows work, what changes is the work style of 200 million people. The impact at this scale cannot be matched by any Agent tool targeting professional users.

Of course, from "usable" to "useful," from "handling simple tasks" to "handling complex workflows," Doubao's Agent capabilities still have some way to go. But the direction is right, and the momentum out of the gate is strong.