Baidu unveiled major upgrades to its artificial intelligence models at the Baidu Create 2025 AI Developer Conference on April 25, pairing technical improvements with sweeping price cuts. Yet it was CEO Robin Li’s critique of rival DeepSeek that captured the spotlight, exposing the intensifying rivalry shaping China’s AI landscape.
Turbocharged models at a fraction of the price
The Chinese tech giant introduced ERNIE 4.5 Turbo and ERNIE X1 Turbo, enhanced versions of its foundational models first released in March. The upgrades promise faster response times, broader multimodal capabilities—including the ability to understand and generate images and audio—and improved tool invocation.
Baidu also announced sharp price reductions. The API cost of ERNIE 4.5 Turbo has been slashed by 80%, priced at just CNY0.8 (US$0.11) per million tokens for input and CNY3.2 for output—only 40% of the price of DeepSeek V3, according to Baidu.
ERNIE X1 Turbo has seen a 50% price drop, costing CNY1 (input) and CNY4 (output) per million tokens, roughly a quarter of DeepSeek R1’s price. Both models are now freely accessible via ERNIE Bot.
Robin Li calls out DeepSeek’s shortcomings
In an unusually direct keynote address, Robin Li praised DeepSeek for advancing AI applications but sharply criticized its performance and cost. According to reports from Chinese media, including Blue Whale News and Sina.com, Li described DeepSeek as “slow and expensive” compared with other mainstream large models in China.
He also pointed to critical limitations: DeepSeek remains text-only, lacking the multimodal capabilities increasingly demanded by Baidu Cloud customers, and suffers from a high “hallucination rate,” making it unreliable for complex applications.
“Without practical applications, neither advanced chips nor sophisticated models hold value,” Li declared, emphasizing that real-world deployment, not just technical prowess, will define AI leadership.
Pushing applications to the forefront
Reflecting this shift toward practical use, Baidu unveiled several new AI-powered services. These include Xinxiang, a multi-agent collaboration platform for complex task-solving; Cangzhou OS, a content operating system combining Baidu Wenku and Baidu Drive; and AI Note, described as the industry’s first multimodal AI note-taking tool. The company also introduced highly realistic AI-generated digital humans with one-click creation capabilities.
Competition intensifies in China’s AI sector
Li’s remarks quickly ignited debate across Chinese social media, highlighting the rising tensions as Chinese tech firms race to expand their AI capabilities.
By combining aggressive price cuts with technical upgrades, Baidu appears to be positioning itself not only as a technology leader but also as a cost-effective platform for developers and businesses alike.
With multimodal AI emerging as the new standard, Baidu’s strategy signals a broader shift: in China’s AI race, technical excellence alone may no longer be enough. Speed, price, and real-world application are fast becoming the metrics that matter most.
Article edited by Jerry Chen