in the meantime DeepSeek doing well by doing good, the opposite of evil
interesting that Meta saw fit / merit to purchase Manus, the user-interface AI agent that in truth uses Grok as LLM, and for $2 billion. Am wondering how many US companies wish to buy application software from China Chinese to jump the gulf that American Chinese cannot leap :0)))
scmp.com
China’s DeepSeek adds advanced ‘thinking’ feature to chatbot amid buzz over next model
Hangzhou-based AI start-up installs ‘interleaved thinking’ as monthly users surge 90 per cent to nearly 131.5 million
Vincent Chow
Published: 9:00am, 6 Jan 2026
Chinese artificial intelligence start-up DeepSeek has updated the user interface of its flagship chatbot by adding an advanced “thinking” feature, as its app recorded a massive 90 per cent jump in monthly active users (MAU) in December.
The upgrades, including a deep research mode similar to OpenAI’s namesake product, marked a rare improvement on the consumer side for the Hangzhou-based firm at a time when anticipation was building over its next-generation models, analysts said.
DeepSeek’s website and mobile app have both started offering an advanced search function called “interleaved thinking”, in which the chatbot performs multi-step research with “thinking” interspersed throughout the process.
Previously, the chatbot would “think” through a prompt provided by a user before generating a response. The “interleaved” allows the model to “think” between each action, strengthening its research capabilities.
This process is integral to powerful research tools like OpenAI’s Deep Research, which can handle complex queries. For instance, after opening a webpage, the model may need to “think” about the credibility of the information it reads before deciding whether to consult another document to verify the findings.
Other Chinese start-ups, including Kimi AI model developer Moonshot AI, have recently added a “deep research” mode to their chatbots. Photo: Shutterstock
A test by the Post showed that DeepSeek’s chatbot does not always deploy this “deep research” mode unless prompted with complicated queries. Once the feature is deployed, the user is able to see each step of the “interleaved thinking” process.
Other Chinese start-ups, including Moonshot AI and MiniMax, have recently added similar features to their chatbots.
Another addition to DeepSeek’s chatbot is a timeline interface that allows users to toggle between their questions, a function seen in only a few other chatbots such as Grok, the flagship product of Elon Musk’s start-up xAI.
DeepSeek did not officially announce the upgrades, but “interleaved thinking” was introduced as a capability in its latest model, V3.2, released in early December, which the company said offered reasoning power rivalling Google DeepMind’s Gemini 3 Pro.
According to AI product tracker Aicpb.com, DeepSeek’s app was the fastest-growing AI app in China last month, with MAU rising to nearly 131.5 million.
The sharp jump reversed months of declining visitors as DeepSeek, known for its focus on core research, trailed domestic rivals in improving user experience over the course of 2025, said Aicpb.com founder Li Bangzhu.
For example, while DeepSeek’s chatbot still lacks multimodal support, ByteDance’s Doubao, the most popular AI product in China, can already process and generate text, voice, image and video.
The interface upgrades could lead to a further increase in DeepSeek’s users, said Delta Wu, head of Chinese AI research firm Unique Research. DeepSeek’s groundbreaking R1 model, released a year ago, drew widespread attention for being the first AI chatbot to reveal its thinking process to users, a feature later adopted by competitors.
“However, I think DeepSeek’s product is still fundamentally a testing ground for its models, where feedback is collected and iterations are made,” said Wu. “So far, they have shown that they don’t much care about their number of end-users, with the core mission still being achieving artificial general intelligence.”
Industry expectations are high that DeepSeek will release its next major model in the run-up to the Spring Festival holiday in mid-February. The company released R1 on the eve of the holiday last year. |
