China, with its vast size and heavily shielded digital landscape, has created a significant advantage for its domestic companies.
By enforcing strict censorship and policies that limit the use of non-approved foreign products, the country effectively secures a market of over a billion internet users for local players.
In an era where AI dominates global attention—largely propelled by Western influence and OpenAI—China has been quick to respond with its own advancements.
Tech giant Baidu has positioned itself as one of the primary rivals of the West, and it began by rivaling OpenAI's GPT-4 with its own AI model it calls ERNIE.
Following the hype of GPT-4.5 that's even more powerful that its predecessor, and after another Chinese company, DeepSeek, created the reasoning model DeepSeek-R1, Baidu plays its card, and unveils 'Ernie 4.5' and 'Ernie X1.'
Making it freely available to individual users via the ERNIE Bot platform—well ahead of the originally planned April 1 release, the strategic move highlights Baidu’s dedication to democratizing cutting-edge AI technology while further expanding its ecosystem of AI-driven tools and services.
The first, which is ERNIE 4.5, is literally a next-generation multimodal foundation model.
What it does, is capable of seamlessly integrating text, images, audio, and video data. It delivers enhanced language understanding, generation, reasoning, and memory while tackling challenges like hallucination prevention and logical consistency.
Boasting both the 'FlashMask' Dynamic Attention Masking and Heterogeneous Multimodal Mixture-of-Experts,' this ERNIE 4.5 is able to perform better than OpenAI's standard GPT-4o and matches capabilities with the newer GPT-4.5.
And all that while operating at just 1% of the cost.
ERNIE 4.5's text capabilities also surpassed that of DeepSeek-V3.
Baidu attributes the significant enhancements in ERNIE 4.5's capabilities to several key technologies including "FlashMask" Dynamic Attention Masking, Heterogeneous Multimodal Mixture-of-Experts, Spatiotemporal Representation Compression, Knowledge-Centric Training Data Construction, Self-feedback Enhanced Post-Training.
As for ERNIE X1, Baidu designed it to be a deep-thinking reasoning model.
Touted as the company's first multimodal reasoning model, the AI is designed for deep-thinking tasks that require advanced logic and precision.
For starters, it excels is Chinese knowledge, manuscript writing, complex calculations, and tool-assisted tasks like image analysis, code interpretation and academic research.
The AI boasts groundbreaking methodologies, such as the Progressive Reinforcement Learning Method, End-to-End Training Approach Integrating Chains of Thought and Action, and A Unified Multi-Faceted Reward System, are able to enhance its reasoning capabilities to whole different level.
While Baidu did not provide benchmark results for its the reasoning model Ernie X1, but the company said that it "delivers performance on par with DeepSeek R1 at only half the price."
In a press release, Baidu said that:
Baidu’s latest AI advancements put increasing pressure on Western AI companies, which continue to charge premium prices for comparable capabilities.
This shift traces back to the "DeepSeek Moment" in January 2025, when DeepSeek proved that Chinese AI labs could rival Western firms with far fewer resources.
Recognizing the competitive threat, Anthropic and OpenAI have urged the U.S. government to regulate Chinese AI development, citing national security concerns. However, the rise of open-source Chinese models also poses a direct challenge to their commercial API businesses.
As China’s leading search giant, Baidu is moving fast in the AI race, ramping up AI development to solidify its position in the global market.
Further reading: China Races To Put DeepSeek Into Everything: How The U.S. Failed To See The First AI War Coming