Baidu expands AI ecosystem with new model, marketplace and chip roadmap

The shift to intelligent organizations

Baidu CEO Robin Li recently introduced the concept of the 'intelligent organization' as the next phase of the AI era. He argues that companies must move away from traditional hierarchies to fully unlock the productivity of new technologies. Li compared AI to electricity, suggesting that organizations need an AI-native approach to function effectively in the modern economy.

To achieve this, Li proposed three specific principles for business evolution. Organizations should break down role boundaries so that people focus on tasks rather than fixed positions. He suggested that roles will eventually converge into two categories: 'Builders' who create new products and 'Sellers' who build emotional bonds and trust with customers.

The company also emphasizes capturing tacit knowledge as a corporate asset that competitors cannot easily replicate. By turning practical experience into reusable skills, businesses can create a shared capability for the entire workforce. Baidu claims that these principles have already helped their internal agent activity grow from 5,000 to 37,000 daily active agents over the last year.

New models and marketplace tools

A major part of this update is the launch of ERNIE 5.5 Flash Preview, a foundational model designed to balance intelligence with reasoning efficiency. It is optimized using Baidu's proprietary Recursive Self-Improvement technology to reduce costs while maintaining high performance. The model's API is now available on the Qianfan platform, and the company plans to open-source the model weights before November 25.

Baidu is also introducing Token Exchange, which serves as a neutral marketplace for AI model services. This platform allows customers to compare different models side by side based on clear pricing and performance data. It aims to help high-quality providers reach more customers while providing a trusted environment for businesses to find and purchase model tokens.

The marketplace relies on two pillars of trust: verification and credit rating. Baidu verifies resources by comparing model fingerprints and testing their capabilities and availability. In its first month, Token Exchange integrated over 100 model resources and processed more than 50 million calls daily, handling over 700 billion tokens.

Specialized agents and video tools

Baidu is also upgrading several specific agents to help with creative and operational workflows. The DuMate agent now serves as a unified entry point for Baidu's suite of specialized tools. During its development, the DuMate team iterated over 160 times in six months, with 90% of its code and 50% of its external communication materials generated by the agent itself.

For creative work, the company launched the ERNIE Wizshot Micro-Expression model, which is specialized in high-expression video creation. This tool can process text, images, and audio simultaneously to produce short animated series or podcasts. It is designed to lead the industry in production costs, reportedly achieving these results at only half the typical cost.

Another tool, KOOKO, is a global platform for documents and creative workflows that integrates professional databases and academic papers. It helps users create high-quality outputs like research reports and websites by leveraging 1.8 billion professional resources. Additionally, the Famou agent is being used to drive continuous optimization in live production systems across 25 of China's industrial sectors.

Hardware and autonomous driving

The underlying infrastructure for these AI services includes new cloud and chip technologies. Baidu Intelligent Cloud unveiled the Tianchi 1024 Card Super Node, which uses a fully optical interconnect for high-speed data transmission. This hardware supports the expansion of the company's AI services and its autonomous driving initiatives.

The company also shared a roadmap for its Kunlun Core AI chips. The M100 inference chip has already shipped, while the flagship M300 is expected to arrive in 2027. These chips are designed to deliver significant advancements in computing power and memory. Li Yanhong noted that inference will be the primary driver of long-term growth in the AI chip market.

Finally, Baidu's autonomous ride-hailing service, Apollo Go, continues to expand globally. It currently operates in 28 cities and has completed over 350 million kilometers of autonomous driving. The service has already launched fully driverless commercial operations in Abu Dhabi and Dubai, and it recently introduced the world's first driverless car rental model in Hainan.

}