The Four AI Tigers Enter the Brutal Competition

"Tough as Daoge"

StepFun finally has news.

Yin Qi returns, Jie Tang resurfaces. Even though Chuan Wang is still talking tough about Baichuan having 3 billion RMB in cash, the "Six Little Tigers" of AI have completely become the "Four Little Tigers." Pure founder-driven projects, officially entering the stunt competition phase.

MiniMax and Zhipu AI have both gone public, taking the most mainstream path. Moonshot AI responded by announcing a $500 million funding round — a real stunt, trying to say: I'm not going public, but I'm raising about as much as you did.

So what's StepFun up to?

I originally thought StepFun just occasionally released a model, did a little product demo here and there. Didn't expect them to pull off something this big: Yin Qi officially announced his return, over 5 billion RMB in Series B+ funding — more than what MiniMax and Zhipu AI raised in their IPOs, tough as Daoge opening a juice shop at the Shenyang Street ruins.

Here's my take: The foundation model competition has pure and simply entered the lighter-biting stage.

The four surviving foundation model companies are like Knife Bro, Tiger Bro, Weird Pigeon, and Feng Bro in the post-Kuaishou era. Desperately pulling stunts to prove they still exist.

Releasing a product is Tiger Bro biting a lighter — the crowd decides whether it's a bad stunt or just mediocre. Funding rounds and IPOs are the hardcore stunts, like diving into a Harbin ice hole in winter.

I hereby announce the commencement of the First Foundation Model Company Stunt Competition.

Now approaching the arena: StepFun, led by Yin Qi.

Yin Qi learned a lesson in the AI 1.0 era: "All glory that can't form a closed loop is temporary." Publishing papers doesn't directly convert to revenue; tech faith needs to be paired with concrete commercial value. Business model is the best model.

Today's foundation model companies face the exact same problem — both 2B and 2C are brutally hard.

The 2B dilemma: if an enterprise wants AI capabilities, why go to a model company instead of a full-service platform like Alibaba Cloud or Volcano Engine?

The 2C dilemma: can't outcompete Doubao, can't replicate Manus.

StepFun's desktop companion is genuinely useful, but it's Tiger Bro biting a lighter — a small stunt, not yet in the major leagues.

To close the loop, you have to think hardware-software integration. The most important hardware-software integration ordinary people can access? Phones and cars.

StepFun likes to say they have a new approach — not 2B, not 2C, not 2G.

StepFun goes 2 Big Brother. Find a big brother, bind to a specific scenario, do hardware-software integration, and go 2C together.

StepFun contributes the tech — over 30 multimodal models in its matrix. Qianli contributes the terminals — hardware-software R&D and engineering capabilities. Geely contributes manufacturing capacity and distribution scale.

The commercial path is clear: first deeply serve Geely, then export smart vehicle solutions to other automakers. Another detail I've seen: StepFun has also reached partnerships in the phone space with OPPO, HONOR, and others.

So Yin Qi isn't a tech genius or a business genius — he's pure connection genius.

He's strung it all together: Megvii's computer vision, hardware-software integration, StepFun's foundation model capabilities, Geely plus Qianli's cars and autonomous driving.

We know Feng Bro's special power is connection, so Yin Qi is the Feng Bro of AI, the Connection Immortal.

The AI industry is kindergarten level. MiniMax and Zhipu AI combined do less than $100 million in annual revenue. Now they're jumping into the mature automotive industry, competing on the same stage as Tesla with tens of billions in annual revenue, Xiaomi with tens of billions in RMB revenue — straight from kindergarten to the school of hard knocks.

Compare the other foundation model companies: Zhipu AI and MiniMax chose the IPO mainstream route. Moonshot AI is raising funds and holding out for something bigger. StepFun chose the heaviest, most complex full-stack closed loop: two wheels + four wheels + two legs — full coverage of phones, cars, and robots.

The man's made plenty of promises too: Qianli will build "China's best autonomous driving system," over 1 million vehicles by 2026, StepFun's models will enter the world top tier, plus incubating various hardware.

No comment from me, but I love the confidence. Doing AI requires ambitious goals — if the goals aren't big enough, how are the people supposed to live?

First you had Li Auto and Xpeng Motors desperately emphasizing they're AI companies. Now you have AI companies saying they're going to build cars. The essence of the world is one giant stitched-together little burger 🍔

What does Jarvis mean? To wrap up the worldview: foundation model companies are genuinely hard to close the loop on.

Everyone kept comparing the Six Little Tigers to the Four Little Dragons, and the more you compare, the more you see the resemblance. Being technically brilliant doesn't mean commercial closure; a model that can reason doesn't mean a business model that works.

So what to do? Really just two stories.

One story is to only do foundation models, firmly believing AGI can be achieved through large language models. The core belief: as long as the model is strong enough, application scenarios will naturally follow. I actually believe this, because I genuinely bought MiniMax stock, and as of now I'm down almost 400 HKD.

The other story is that you can't only do foundation models. I don't believe large language models alone lead to a bright future. Instead, you need to combine foundation models with specific business scenarios, hardware-software integration, combined with cars and phones. Just these two paths.

Finally, callback to Immortal Yin Qi.

In the era of the AI Four Little Dragons, Yin Qi was already playing in the major leagues. Then he moved to automotive intelligence — Qianli plus Geely — still major leagues. Then pushing for StepFun's founding, now stepping in as chairman, treating every tech wave of the past decade as an achievement to unlock.

The Major Leagues Immortal, always active at the front lines of the table, always holding good cards, always having big brothers deal him more.

Will the loop close this time? Blessings to StepFun, tough as Daoge opening that juice shop.

(Article illustrations generated by NanoBanana, with writing assistance from Claude Code)