WeChat DeepSeek entry has been connected, have you been selected for the internal test?

Author: Bai Jiao, Heng Yu, from Aofeisi Temple, Quantum Bit

Image source: Generated by Wujie AI

Open your eyes and confirm via WeChat:

WeChat has officially integrated DeepSeek-R1—the full-powered version!

The process is simple: Open WeChat — Click the search bar — Select "AI Search" in the middle of the page — Choose Deep Thinking mode — Start using it~

The generated results come with a social feature—shareable to friends or postable to Moments.

Previously, many users had requested an easy way to share DeepSeek answers instantly, but this wish remained unfulfilled. Before last night, almost all attempts to share DeepSeek-generated content involved screenshots (and due to its visible reasoning process, these screenshots were especially long).

Now, WeChat has finally made this happen.

The official account "Consensus Crusher" calculated overnight that WeChat currently supports 50 million to 100 million concurrent online users (concurrent, not DAU).

This number basically meets the initial demand from WeChat's first batch of DeepSeek users and exceeds ChatGPT’s concurrent user count.

Although still in gray-scale testing, this is clearly Tencent’s continued full embrace of DeepSeek—and now it’s truly ready for hands-on use.

Good news—the Quantum Bit editorial team was among the first to receive gray-scale access.

Time to celebrate—shall we start real-world testing? ——

Testing WeChat’s Integration with DeepSeek

Currently, WeChat’s AI search offers two modes:

• Quick Answer: Generates concise responses, suitable for general, simple questions.

• Deep Thinking: Powered by DeepSeek-R1, supports extended reasoning and comprehensive answers, delivering more detailed results.

To fully experience R1, select the "Deep Thinking" mode.

First, let’s check if it supports basic web-connected mode.

In recent days, Unitree Robotics has once again amazed everyone with new dance moves—updating their dancing video, even more agile and stunning than the Spring Festival Gala performance, leaving foreigners stunned.

Ask WeChat: What dance is in Unitree Robotics’ latest dancing video?

The answer received:

Its response was smooth, with fast output. It first showed its reasoning process, determined through online search that this was a development from February 14, 2025, then, considering the user’s interest in robots, provided additional context, ultimately arriving at the following answer.

There’s a small bug here—the thinking process and final answer aren’t clearly separated, and formatting didn’t render properly. But this is likely due to instability during gray-scale testing; it should stabilize later.

From the 10 references attached, the first seven originate from official accounts. This proves that behind this AI search lies the massive content ecosystem of WeChat.

Alright, basic web-connected reasoning works—now let’s increase the difficulty.

Previously, people used DeepSeek mainly for fun tasks like fortune-telling or creative character writing.

So then:

Please write a ci poem in the style of Xin Qiji on the theme of WeChat integrating DeepSeek.

After completing its thinking, it hides the thought process and shows the final answer, which can be expanded upon clicking.

The final output is as follows:

While generating the poem, it also provided annotations, containing many hidden nuances.

The first half represents the significance of the event—"wisdom opens the Milky Way"—carrying dual meanings; the second half reveals functional features, such as accessing data from WeChat’s ecosystem.

After answering, it recommends related questions or prompts “ask another question.”

Thus, by clicking a suggested question or asking a follow-up, the conversation can continue indefinitely (though it does not support contextual continuity across multiple queries).

And as long as you don’t exit, all Q&A will appear within the same interface.

Overall, due to gray-scale testing limitations and the current interface (mobile-only), there are areas needing improvement.

For example, only mobile search is supported, and shared content sent to friends or posted to Moments cannot be viewed on desktop.

Besides that, the answers aren’t as diverse as the official version—likely due to token output limits—and lack features available in the official release, such as retrieving past conversations or maintaining context across multiple turns.

Thus, WeChat’s integration with DeepSeek currently presents a short, fast, and efficient search method, fitting well with WeChat’s instant, private social nature.

While the world struggles over AI entry points…

What effect does WeChat’s integration with DeepSeek produce?

In one sentence:

While global tech giants scramble for dominance in AI access, Tencent’s seemingly effortless “ecosystem grafting” has led netizens to jokingly call it “the end of the AI search race.”

Does WeChat deserve such praise? Let’s analyze.

First, based on real-world testing, even when both use DeepSeek-R1, Tencent’s Yuanbao feels faster and smoother than WeChat.

But that’s not important.

What matters is that WeChat’s integration with DeepSeek-R1 provides a zero learning-cost access point for an immeasurably large potential user base.

Thus, 1.67 billion WeChat users (data from internet sources, as of October 2024) can now directly access China’s most powerful reasoning large model without downloading a new app or registering a new account.

Second, and truly the killer advantage: under the premise of offering an easily accessible entry point, this integration deeply fuses R1 with WeChat’s ecosystem data resources and RAG (Retrieval-Augmented Generation) technology.

When a user asks, “How to book tickets to the Forbidden City,” DeepSeek doesn’t just search the web—it uniquely pulls relevant materials from WeChat Official Accounts, Channels, and Mini Programs, and can even invoke service interfaces from the Forbidden City’s official Mini Program. This is Tencent’s exclusive ecosystem advantage.

Finally, answers are presented clearly with emojis, making them easy to read and key points easy to grasp—adding a human touch.

As we’ve previously noted, Tencent’s embrace of DeepSeek is broader and more comprehensive.

Prior to this, Tencent Cloud, the large-model knowledge application platform Knowledge Engine under Tencent Cloud, the AI workspace ima, and flagship AI app Yuanbao—all members of the Tencent family—have successively integrated DeepSeek-R1.

Such open-source success and Tencent’s openness lead to one outcome:

Users can access powerful AI with the lowest learning cost, in the most convenient way, and on the widest scale—right inside their phones.

The official account “Consensus Crusher” called this move “a turning point for consumer-facing AI applications” in its latest post.

It also revealed some insights:

△ Screenshot from the article “WeChat + DeepSeek: A Turning Point for 2C Applications” by “Consensus Crusher”

Super apps from the mobile internet era continue today to leverage their advantages—and retain the mindset to create super apps.

In short, starting today, China’s most frequently used app, its strongest reasoning model, and high-quality ecosystem data resources have been seamlessly combined by WeChat and DeepSeek.

One More Thing

Yesterday, Perplexity launched Deep Research for free.

In terms of performance, tested against Humanity’s Last Exam—a popular benchmark for AI reasoning—it achieved a 20.5% accuracy rate.

Outperforming Gemini Thinking, o3-mini, o1, DeepSeek-R1, and many other mainstream models.

In response, Perplexity’s CEO publicly thanked DeepSeek:

Because DeepSeek is open-source, cheap, and fast.

But Altman isn’t staying quiet—this morning, he took to Twitter, directly targeting Perplexity’s CEO.

He said:

ChatGPT (4o) is the best search product on the web!

But Now?

DeepSeek says nothing. WeChat smiles silently.