Loading Studio Assets...
Baidu — China's original AI-search giant — has released ERNIE 5.0, a 2.4-trillion-parameter omnimodal flagship. The market noticed: Baidu's stock climbed to its highest level in nearly three years following the launch.
"Omnimodal" is the pitch: ERNIE 5.0 is built to handle text, images, audio, and video in a unified model rather than bolting separate systems together. That's the same direction OpenAI, Google, and others are pushing — a single model that natively understands and generates across formats.
The most strategically important detail isn't the parameter count — it's that ERNIE 5.0 was trained on Baidu's own Kunlun silicon. In a world where US export controls limit China's access to Nvidia's top chips, training a 2.4T-parameter frontier model on homegrown accelerators is a statement of self-reliance.
If China can train frontier models without Nvidia, the leverage of export controls weakens over time. ERNIE 5.0 is a data point that the domestic-silicon path is viable.
Baidu also moved to open-source its latest ERNIE models as competition intensified — a notable shift for a company that long kept its best models proprietary. It's a response to the gravitational pull of China's open-weight ecosystem, where DeepSeek, Qwen, and Kimi set the tempo.
Inside China, ERNIE Bot is one of five AI search/assistant platforms — alongside Alibaba's Quark, Tencent's Yuanbao, ByteDance's Doubao, and Moonshot's Kimi — that together serve over 900 million monthly active users. It's a massive domestic market largely walled off from Western models.
ERNIE 5.0 matters less as "another big model" and more as proof that China's full-stack AI — homegrown chips plus frontier models — is coming together.
Brandomize helps global brands understand the fast-moving AI landscape, East and West. If China's AI stack affects your market, we can help you make sense of it.
We help founders, brands, and local businesses turn modern tech into measurable revenue and standout brand identity.
As the internet drowns in recursive synthetic sludge, artificial intelligence is eating its own tail—triggering irreversible model collapse and epistemic decay.
OpenAI CFO Sarah Friar argues proprietary models beat open source on total cost of ownership, citing an 80% Luna price cut and useful intelligence per dollar.