China’s ‘AI Tigers’ Pounce: A New Wave of Models Challenges US Dominance

China’s artificial intelligence race is picking up speed again. 

From startups to tech giants, Chinese companies are rolling out new AI models at a rapid pace, signaling that the gap with US rivals may be narrowing faster than many expected.

Over the past few months, several Chinese firms have unveiled new AI upgrades, adding to the momentum that began last year with DeepSeek’s surprise release of its R1 model. That launch drew global attention by offering performance comparable to ChatGPT at far lower costs, raising questions about whether US export controls were slowing China’s AI ambitions at all.

On Tuesday, the Beijing-based startup Moonshot AI announced Kimi K2.5. This multimodal powerhouse can process text, images, and video simultaneously. Perhaps more impressive is its “agent swarm” feature, which allows a user to deploy up to 100 sub-agents to tackle massive workflows in parallel.

Hours earlier, e-commerce titan Alibaba launched Qwen3-Max-Thinking. The company claims this model didn’t just compete, it dominated. According to Alibaba, the model outperformed major US rivals on “Humanity’s Last Exam,” a notoriously difficult benchmark test.

“It seems that, for these Chinese start-ups, it was just a matter of getting access to capital,” said Kyle Chan, a fellow at the Brookings Institution, in a report by SCMP. “I wouldn’t have guessed that Chinese AI companies would continue to keep pace with their US peers as recently as a month or two ago.”

Other firms are moving just as quickly. Z.ai recently released a free version of its GLM 4.7 model, but later restricted new sign-ups after demand strained its computing capacity. Baidu, meanwhile, saw its Hong Kong-listed shares climb to their highest level in nearly three years after unveiling its latest Ernie 5.0 model.

Open Source models gain global traction

While American firms like OpenAI and Anthropic largely keep their “secret sauce” behind paywalls, Chinese firms are leaning into an open-source strategy. By releasing the “weights” of their models, they allow developers worldwide to customize the tech for free or at a very low cost.

The price difference is staggering. Artificial Analysis, cited by SCMP, estimates that Moonshot’s K2.5 is more than four times cheaper to run than the leading models from OpenAI. This affordability is already winning over Western tech giants. Airbnb recently swapped its customer service AI for Alibaba’s Qwen, while Pinterest has integrated Chinese models to power its recommendation engine.

Navigating the ‘Silicon noose’

The surge in Chinese innovation comes despite US restrictions on the advanced Nvidia chips usually required to train these massive systems. Instead of slowing down, Chinese engineers are getting creative, maximizing “Mixture-of-Experts” architectures to do more with less computing power.

Some are even looking inward for hardware. Zhipu AI recently launched GLM-Image, claiming it is the first domestic model fully trained entirely on Chinese-made chips. Even Google DeepMind CEO Demis Hassabis acknowledged the closing window, telling CNBC that China’s models might only be “months” behind those in the US.

Beyond the screen: Robots and red envelopes

The competition isn’t staying digital. Tencent is using the upcoming Lunar New Year to bake AI into daily life, promising $140 million in cash awards through its Yuanbao AI app. Meanwhile, XPeng founder He Xiaopeng recently demonstrated a humanoid robot named IRON that moves with such human-like precision he had to cut the robot’s leg open on stage to prove it wasn’t a person in a suit.

As these “AI Tigers” continue to raise billions — with Moonshot recently hitting almost $4.8 billion valuation — the focus is shifting from “Can we build it?” to “How fast can we weave it into the fabric of the global economy?”

Also check out this eWeek guide on the top eight things you should never paste into an AI chatbot to protect your privacy and security.