DeepSeek, until recently a little-known Chinese AI startup, shook up the U.S. tech industry over the weekend when it unveiled a cutting-edge large-language AI model that could compete with the big guns like OpenAI, but at a fraction of OpenAI’s budget.
The tech-heavy Nasdaq dropped 3% Monday, and AI chipmaker Nvidia alone lost almost $600 billion as DeepSeek’s cheaper and similarly capable model led investors to question the amount of capital that has been poured into AI development.
But DeepSeek isn’t the only Chinese tech firm to release an AI model in recent weeks, as a slew of Chinese AI players have been rolling out updates ahead of the Lunar New Year on Wednesday, when the country traditionally takes at least a weeklong break. But while it could be a case of companies releasing their best work ahead of a holiday, analysts think the flurry of activity is something quite different—that is, “coordinated psyops to counter U.S. announcements from the past week.”
That U.S. announcement was Trump’s presentation of a $500 billion project called Stargate that’s aimed at building AI infrastructure in the U.S.—an announcement that comes on the heels of months of AI chip export bans announced under former President Joe Biden.
“Necessity is the mother of invention, so the chip export control bans may have caused this challenge,” said Ray Wang, principal analyst and CEO at the Silicon Valley–based tech research and advisory firm Constellation Research.
“This is obviously a psyops. You can’t get to AGI [artificial general intelligence] this way. It’s reverse engineering for efficiency,” Wang added, in reference to DeepSeek’s role as a low-budget competitor to the likes of OpenAI. According to Wang, despite all the buzz around DeepSeek, AI models will keep getting more demanding and complex over time, which will require large amounts of expensive computing power.
Another analyst, at IDC, a market intelligence firm, holds a similar view and thinks China wants to show that it is still a force to be reckoned with when it comes to tech.
“It’s hard to say if they timed it for Chinese New Year, however with the new presidency in the U.S. and the announcement of huge AI investments developing AI infrastructure through Stargate, et cetera, there is a need for China to reinforce its position in the global tech industry,” said Deepika Giri, head of AI research at IDC APAC.
Recently released Chinese AI models:
MiniMax
This Chinese startup launched a new series of open-source models two weeks ago under the name MiniMax-01. That family includes a general purpose foundational model, the MiniMax-Text-01, and the visual multimodal model MiniMax-VL-01.
The developers claim the MiniMax-01, which is 456 billion parameters in size, outperforms Google’s recently released Gemini 2.0 Flash on some benchmarks like MMLU and SimpleQA. MMLU stands for massive multitask language understanding and is a benchmark used for evaluating large language models across a wide range of tasks. SimpleQA measures a large language model’s ability to answer short fact-seeking questions.
Besides the MiniMax-01, the startup, backed by Tencent and Alibaba, also released an audio generator labeled T2A-01-HD. The company claims that the application can generate “premium-quality output” from just 10 seconds of audio input, and can capture voice characteristics, speech patterns, and emotional nuances.
Qwen
Qwen, also known as Tongyi Qianwen, is a large language model backed by Alibaba. The company claimed in May of last year that Qwen has been adopted by over 90,000 corporate clients in areas ranging from consumer electronics to automotives to online games.
On Monday, the Qwen team released Qwen2.5-VL, which can perform various types of image and text analysis tasks as well as interact with software either on a PC or smartphone.
A video circulating on X posted by Philipp Schmid, technical lead at Hugging Face, shows Qwen2.5-VL launching the Booking.com app on Android and then booking a flight from Chongqing to Beijing.
The Qwen team claims the Qwen2.5-VL beats GPT-4o when compared on video benchmarks.
Tencent
Tencent is one of China’s biggest tech firms and the owner of WeChat, the super app that has 1.3 billion monthly users. But besides the app, Tencent is also a major player in the video games industry with stakes in companies like Supercell, Riot, and Epic Games.
Tencent released the Hunyuan3D-2.0 last week, an update of its open-source Hunyuan AI model that could revolutionize the video games industry. Highly skilled artists can often take days or even weeks to create 3D models and characters in video games, and Tencent’s newer version is expected to make it easier and faster for these developers to produce them.
“Creating high-quality 3D assets is a time-intensive process for artists, making automatic generation a long-term goal for researchers,” the company wrote in a technical report.
The Hunyuan3D-2.0 includes two foundation components: a large-scale shape generation model, the Hunyuan3D-DiT, and a large-scale texture synthesis model called Hunyuan3D-paint.
The team claims the Hunyuan3D application outperforms previous state-of-the-art models including both open-source and closed-source models when it comes to parameters like geometry details and texture quality.
