DeepSeek, Kimi, and Qwen compared on work tasks: a test of five Chinese AI assistants

In a new video comparison, the editorial team put five Chinese AI assistants — Doubao, Qwen, Yuanbao, Kimi and DeepSeek — through typical work tasks that usually require not code, but everyday intellectual routine. The focus was on text summarization, data analysis, industry insights, and business writing.

How the test was conducted

The starting point of this comparison is simple: AI already handles a significant portion of text work that used to fall entirely on humans. That's why the authors didn't measure abstract benchmarks, but instead chose a more practical scenario — observing how different assistants behave in tasks resembling real requests from an editor, analyst, manager, or researcher. This approach matters because users typically don't need "the smartest" bot in general, but a tool that saves time here and now.

"AI can already solve most of the text tasks I encounter at work."

The test included five notable players in the Chinese market: Doubao, Qwen, Yuanbao, Kimi and DeepSeek. They were evaluated under identical or closely similar conditions to see not only the quality of the response, but also how the model maintains structure, handles phrasing, and deals with tasks where the goal is not simply to continue a text, but to extract value from it. This comparison is more about practical usability than about impressive demonstrations of capabilities.

What skills were tested

The set of tasks reflects the most common office use case for AI: the user brings a large volume of text or data and wants to quickly get a clear result. This is precisely where differences between models are most visible, because a polished presentation alone is not enough — accuracy, compression, logic, and the ability to maintain the goal of the response through to the end all matter. This is the type of work where a mistake is immediately apparent and quickly affects the quality of the final document.

• Condense long material into a brief and clear summary
• Break down data and highlight key figures or anomalies
• Formulate industry insights from the given inputs
• Write text in the required style and with a clear structure

Testing these particular skills shows how suitable an assistant is as a working layer on top of everyday processes. If a model summarizes well but gets confused with numbers, it's hard to trust it with analytical notes. If it writes confidently but can't draw industry conclusions, it remains more of a draft generator. That's why such tests are useful not only for choosing a "winner," but also for understanding the specialization of each tool. It is at this level that the decision is usually made whether to include a model in a permanent working stack.

Where practical differences emerge

The main value of such a comparison lies in moving AI from the wow-effect mode into the utility mode. For businesses and individual specialists, the question is no longer whether a model can answer questions at all. The question is different: can you hand it raw material, quickly get a first pass, and then spend less time on edits. This is exactly where real differences emerge — who maintains context better, who handles data more carefully, and who writes more coherent and publication-ready text.

In the Chinese market, such assistants are becoming more and more numerous, and competition is shifting from general promises to the quality of specific work. Users compare not the number of parameters or the loudness of marketing, but how many steps remain after the model's response. If you have to completely rewrite the text after an AI, its usefulness quickly disappears. But if it produces a clear structure, doesn't lose important details, and saves at least half the time on content preparation, that's already a serious advantage in real work.

What this means

The AI assistant market is maturing: models are increasingly evaluated not by impressive demos but by how they perform in ordinary work tasks. For the user, this is a useful guide: the choice should not be the "best AI in general," but an assistant suited to a specific type of work — summaries, numbers, industry analysis, or draft writing. And the more such comparisons are built around everyday routine, the easier it is to separate a convenient tool from a merely famous name.