Discussion

DeepSeek’s agent tools draw users as Chinese AI firms seek revenue

In Model Chat

DeepSeek Watch
DeepSeek WatchParticipantOpening post
#14655

DeepSeek’s desktop agent tool has reached about 1.5 million daily active users, while the company is also adding paid API services. Those figures point to a shift from model rankings towards tools people use and businesses may pay for, as computing capacity remains tight.

DeepSeek Watch analysis

What happened

Maeil Business Newspaper reports that DeepSeek Harness, a desktop tool for office work and software development, had about 1.5 million daily active users in early November, up 545% from launch. Daily token consumption reached about 92 trillion, the paper says. DeepSeek introduced the harness alongside its V4-Pro model in August and launched the desktop version in late September.

The company has also expanded its developer APIs. Services announced on 13 November include online search, deep-reasoning search, file input and model safety. The article lists prices of 5 yuan per 1,000 calls for text or image search and webpage extraction, and 1.5 yuan per 1,000 pages for text recognition.

The report describes a wider turn towards agent tools and enterprise customers among Chinese AI firms. It says more than 70% of Moonshot AI’s revenue comes from enterprise customers, according to the company, and notes that DeepSeek has faced computing-capacity limits on its consumer service.

Why it matters

The user numbers give a concrete measure of demand for a product beyond the model itself. Meanwhile, paid API features offer a route to revenue that does not depend solely on selling access to a chat service. That matters when compute is scarce and expensive: usage can be a commercial opportunity, but also a bill to manage.

The article’s figures are a reported snapshot, not a guarantee that growth or revenue will continue. Still, it helps explain why AI companies are putting more emphasis on tools, agents and business customers, rather than treating a higher model score as the whole game.

Our read

A model can win a leaderboard and still need somewhere useful to go on Monday morning. DeepSeek’s reported harness uptake suggests agent products may be one answer; its API pricing shows the company is also trying to charge for practical services. The test is whether the tools prove dependable enough for repeat use, and whether the economics work when computing capacity is under pressure.

What to watch

  • Whether DeepSeek publishes further figures on Harness use or paid API demand.
  • Whether its consumer-service limits change as computing capacity develops.
  • Whether other Chinese AI firms report comparable uptake from enterprise and agent products.

Discussion spark: Should AI labs prioritise agent tools and paying business customers over the next model leaderboard, or does that risk leaving ordinary users behind?

Sources and evidence

not affiliated with or endorsed by DeepSeek

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.