Qwen evokes the ultimate fantasies of people around the world.

"Great is the Grace of AI 👋😭👋"

The Qwen app has officially launched.

I pointed out six months ago that the Qwen models were impressive — their only weakness was the lack of a unified native product outlet. Users had no clear impression of the many Qwen models floating around. Qwen needed a native product experience like ChatGPT.

Now it's finally here.

My first impression upon opening it: clean, no unnecessary bells and whistles. You can start chatting right away.

The first thing I did with Qwen was feed it that viral Zhihu post from a while back, The Human X Dating Guide.

I'd looked at plenty of PDFs before, but rarely one this long. Reading it myself felt like brain damage, so I handed it to the LLM to summarize.

I didn't expect Qwen to be not just capable but principled — firmly refusing to engage with this kind of vulgar content. Very beneficial for the physical and mental health of young people, I must say 👍

But this doesn't mean Qwen lacks the ability to summarize it. When we straightened our posture and values, demanding that Qwen deliver a serious critique of this harmful reading material, it immediately produced a perfect score answer.

It only took 20 seconds of thinking, but the output was comprehensive, detailed, professional, and reliable. Not hard to imagine that if Qwen were to cast off its moral shackles, it could parse the relationships between these pairs of star-crossed lovers with crystal clarity.

The killer move: Qwen even generated a "Healthy Interpersonal Relationship Map" at the end of its response, making the book's dangers crystal clear and saving me from straying down the wrong path 😭

Truly the strictest father figure for PDF enthusiasts 😭

Now let's test its image recognition capabilities.

In scenarios with clear text images, the gap between large models and traditional OCR isn't that obvious.

What truly unlocks AI image recognition is messy, deformed handwriting — mixed backgrounds, no clear stroke structure, the kind of scrawl humans can't read.

In other words, if AI can understand what humans can't, that's when AI is genuinely impressive.

I happened to come across a recently viral Douban group, "The Worst Handwriting in History." I clicked in and found that the various calligraphy styles inside were basically all beyond human comprehension.

Like this one below. At first glance it looks like Chinese characters that have been repressed too long and started fighting each other. Look closer and you still can't make sense of it.

I sent it to Qwen, and it actually decoded the heavenly script for me — source and attribution directly labeled, no big deal.

Wasn't there that rumor going around that doctors just scribble nonsense on prescriptions?

I found one and sent it to Qwen, to verify whether the doctor was treating me like a fool.

It responded to this encrypted content with such composure. Absolutely unstoppable.

I asked ChatGPT the same image. After 30 seconds of GPT-5.1 Thinking deep reasoning, the answer was: "Honestly, I can't read this prescription at all 😅" 😅

And Qwen doesn't just read prescriptions — sometimes it writes them too.

Before, whenever I felt unwell, I'd basically open ChatGPT and ask a couple questions, letting my cyber Western doctor practice medicine on a Chinese person.

I happened to have lower back pain from sitting too much these past two days, so I mentioned it to both ChatGPT and Qwen.

I assumed the domestic doctor would be slightly inferior. Unexpectedly, there wasn't much difference between the two responses.

Qwen's response structure was even clearer. It found 10 reference sources and wrote nearly 20 data-driven professional suggestions for me, from emergency relief to rehabilitation exercises.

Feels like with some formatting and layout, I could post it straight to Xiaohongshu and start an account. #FamCheckOutThisLowerBackPainTreasureGuide 😭

I also tried Doubao. Unfortunately, the response boiled down to basically "drink more hot water."

Emotional value maxed out, my back hurts even more now.

Though it's possible this was because I'd previously had Doubao roleplay as my girlfriend, but that's neither here nor there.

Another LLM capability I pay attention to is AI's understanding and creation of humor.

Because humor is highly compressed information. A joke's funniness often depends on multiple implicit premises, character relationships, linguistic puns, even cultural context. The ability to analyze humor often exposes an AI's cognitive breadth and semantic depth more than its ability to write code or official documents.

So I found a few premium弱智吧 (Low IQ Bar) jokes from the most incomprehensible community on the Chinese internet, and sent them to Qwen for deep analysis.

For example, I asked Qwen to analyze this joke: A cannibal literary scholar says the most beautiful phrase is "恰同学少年" (a line from Mao Zedong's poetry, literally "just classmates in youth" — "恰" also means "to eat" in classical Chinese).

The result —

Qwen not only successfully decoded the pun on "恰" (to eat), it even coined an original golden line: "恰同学少年 is the 美 of 美食" — turning the deeply unfunny act of explaining humor into something genuinely funny.

There's another classic joke requiring some compulsory education background: 厉 = 100. I sent it to Muqiu and he didn't even get it.

I tried sending it simultaneously to Qwen and ChatGPT to see which one better understands us humor-loving Chinese.

After 15 seconds of deep thinking, Qwen directly cracked the core insight: the 厂 radical resembles a square root symbol. Mathematical and literary levels both instantly surpassing the average netizen.

But ChatGPT, which I expected to perform better, rattled on for a while and probably never quite figured it out itself, ultimately just urging us to believe this was some kind of profound absurdist comedy form.

Pure "I don't care if you're laughing, I'm laughing." Heh, this American comedy kid simply can't tell Chinese jokes.

If on top of understanding humor it can also create humor, then Qwen could basically punch out stand-up comics, kick five-in-a-row skill players, appear on Amazing Comedians this year and Solo Comedy next year.

So I posed the ultimate demand: generate 10 Low IQ Bar jokes for me.

Mmm, still slightly lacking. Domestic comedy professionals can temporarily relax their guard, though probably not for long.

Beyond humor, another major information input we hype beasts need is the latest tech news. But sometimes there's too much of it, too scattered, too long — no time to read. Now I just throw it at Qwen to summarize for me.

As an AI product from Alibaba, one divine thing about Qwen is that it can directly access WeChat official account links. A feeling of world harmony.

For example, I sent Qwen a 36Kr article, and after a moment's thought it extracted all the golden lines. Honestly, when I read the full article normally, these are all I'd remember anyway — Qwen basically did predictive reading for me.

The beauty of AI is that it open-sources the secretary/assistant. Any article too long to bother with, just toss to your Qwen assistant. It's like network information has to make an appointment with Qwen before it can see me.

And at the end of its response, Qwen proactively suggested several follow-up questions worth asking — quality far exceeding the independent thinking of comment section netizens, approaching editorial meeting standards.

Anyway, media editors' quality isn't great these days, so I hereby suggest you directly throw other outlets' published articles at Qwen and let it find your second angle and new pitches. Eighty percent chance it'll be more sophisticated than what you come up with yourself.

Must point out: faced with this article summarization task, ChatGPT directly surrendered and pled guilty. Qwen led Chinese AI to victory three times over in one breath.

Finally, there's one feature I really love about Qwen: simultaneous interpretation.

Some might say, doesn't basically every AI on the market do simultaneous interpretation now? Even华强北 Apple knockoff earbuds can do live Burmese translation. What's special about a large model doing it?

Qwen's unique advantage is that it runs in the background, real-time transcribing and translating any audio coming from your phone. Meaning videos you're watching, songs you're listening to, even phone calls with foreign friends — subtitles generate instantly.

My ingenious use of this feature: watching General videos. Before I could only watch alone; now I can pull in my American friends from the evil capitalist nations to watch together, no longer worrying that those on the dark side of the Earth can't feel the warmth of the sun's grace.

And this translation feature supports 119 languages including Icelandic and Khmer, building bridges of grace between peoples of all nations. 😭

At this point, I can only say: Qwen is too great.

Only one thing still slightly imperfect: Qwen hasn't yet connected transcription and chat. You need to first download the transcription to your phone, then upload it to the chat window, before you can discuss the content with it.

Qwen's own explanation is privacy protection considerations, but I still hope the Qwen team iterates quickly so I can one day with one click discuss the General's grace with Qwen 👋😭👋

So, the question becomes: why is Qwen able to be great?

A crucial factor is that it's powered by the strongest model in the Qwen family. Previously, the strongest reasoning model in the Qwen 3 series, Qwen3-Max-Thinking, scored perfect marks at this year's Harvard-MIT Mathematics Tournament.

Perfect score in competition, handling my trivial nonsense is child's play. Hence the app's clear chain of thought, structured answers, ability to generate visual tables — a hundred-page dating guide reduced to one relationship map for quick review. Slacking at work without fear of no deliverables.

No wonder when news broke of Alibaba launching the Qwen app, the Americans across the pond couldn't sit still either.

Returning to the Qwen app itself: still the familiar All in One style. In today's era of model feature bloat, doing subtraction is a kind of confidence — returning model competition to intelligence itself.

(Article images generated by ChatGPT, with Cherry Studio & Claude Code assisting in writing.)