AI News

Baidu World 2026 Held, Releasing Token Exchange and ERNIE 5.5 Flash Preview

People's DailyUpdated Nov 13 05:51

AI summary

On November 12, Baidu World 2026 was held in Beijing. Baidu proposed that "the next act of artificial intelligence is the intelligent organization" and put forward the "three principles of organizational evolution" for the first time. The conference released the large model trading platform Token Exchange, which focuses on authenticity verification and credit reporting, with more than 100 supply resources onboarded and average daily Token calls exceeding 700 billion; it also released the ERNIE large model 5.5 Flash preview with 280B total parameters and 14B activated per computation, as well as the ERNIE Lightning deep thinking model and the ERNIE Yijing micro-expression large model. Four major agents—Baidu Dazi, Kuku, Baidu Yijing, and Baidu Famou—were upgraded. Baidu AI Cloud released the Tianchi 1024-card super node. Kunlun Chip M100 has been launched, and M300 will be launched next year.

Why read this

AI score 55

Baidu World 2026 concentratedly released the Token Exchange, the ERNIE 5.5 Flash preview, and upgrades to multiple agents, which can be used to observe the implementation pace of its four-layer architecture of chips, cloud, models, and agents.

Topics

Tags

Body · AI translation

On November 12, Baidu World 2026 was held in Beijing under the theme "The Next Act: Intelligent Organizations." At the conference, Baidu put forward the judgment that "the next act of artificial intelligence is the intelligent organization," and proposed the "Three Principles of Organizational Evolution" for the first time. Baidu calls the AI-native organizational system formed through the application of AI technology the "intelligent organization," likening it to a "new factory" using "new electricity" and describing it as an important carrier of productivity in the AI era.

During the conference, Baidu released a series of achievements centered on the four-layer architecture of "chips, cloud, models, and agents": it officially launched the large model trading platform Ciyuansuo, released the preview version of ERNIE 5.5 Flash and two ERNIE application models; four major agents—Baidu Dazi, Kuku, Baidu Yijing, and Baidu Famou—received upgrades; Baidu AI Cloud launched the Tianchi 1024-card super node, Kunlunxin announced a new product roadmap, and Apollo Go accelerated its global operations.

Using AI well and fully hinges on organizational evolution

In 2016, the Baidu World Conference took "AI is the new electricity" as its theme and made the judgment that "the next act of the internet is artificial intelligence." From 2016 to 2026, "AI is the new electricity" has been validated in more and more fields. Over the past decade, Baidu's AI technology has formed an independent technology stack, and the four-layer architecture of "chips, cloud, models, and agents" has gradually taken shape.

Over the past year, Baidu has held exchanges with companies in finance, manufacturing, energy, and other industries, and has often found that Token spending did not grow in tandem with corporate revenue and profit. Why has the application of AI technology not produced the expected economic benefits? Baidu believes that to use "new electricity" well and fully, swapping in a "new motor" is only the starting point; the key lies in redesigning a "new factory."

Upholding the "Three Principles" to build "intelligent organizations"

The first principle is to "break down job boundaries and unleash organizational vitality." Traditional organizations are centered on positions and emphasize matching people to posts; organizations in the AI era will shift to being task-centered. The organization of the future will be a network of tasks; positions will be integrated into two categories: Builders, who are responsible for creating new things, and Sellers, who are responsible for communication and connection between people and for building trust.

The second principle is to "accumulate tacit knowledge and build organizational intelligence." The practical experience accumulated by each employee is an important asset of the enterprise. But only by making employees' tacit knowledge explicit, completing its accumulation on the enterprise platform, and creating a situation of continuous accumulation, on-demand retrieval, and availability to everyone, can true organizational intelligence be formed.

The third principle is to "design closed-loop processes and drive organizational evolution." Baidu proposed a formula for the continuous evolution of organizations: measurable + Looping = continuous evolution. With objective measurement standards, employees and AI can continuously iterate around tasks, measuring once per cycle and then optimizing based on the results. On this basis, enterprises can become "intelligent organizations" with highly automated operations; organizations can become systems that continuously learn and evolve.

Baidu has already applied this set of concepts to itself. At the 2025 Baidu World Conference, Baidu proposed internalizing AI capabilities to build an AI-native organization. Over the past year, the number of daily active agents within Baidu has grown from an average of 5,000 per day to 37,000, an increase of more than sixfold; the number of effective tasks completed by agents each week has grown from 84,000 to more than 1 million; Baidu has built 4,300 Workspaces and accumulated more than 20,000 organizational Skills. In this way, one employee's experience is integrated into the capabilities of the entire organization.

Ciyuansuo makes large model trading open and transparent

Model iteration is accelerating. In the first nine months of this year, more than one new model was released every day, and demand for API calls rose accordingly. A number of token relay stations emerged, yet they do not take responsibility for the authenticity and performance of models or for the stability and response speed of services. The more models there are, the harder it is for enterprises to choose.

To solve this problem, Baidu officially launched the large model trading platform Ciyuansuo, striving to create an exchange for large models. By bringing together good models and model providers, it enables enterprises and even individuals to conveniently find, purchase, and use trustworthy model tokens.

Ciyuansuo focuses on two major capabilities, "authenticity verification" and "credit rating." It conducts resource verification through related technologies such as model fingerprint comparison, capability testing, and performance and availability testing. At the same time, it uses methods such as regular spot checks and dial tests and full-link real-time monitoring of calls to establish credit ratings for model providers and provide customers with neutral, trustworthy, open, and transparent model trading services. Currently, Ciyuansuo has more than 100 supply resources onboarded, with average daily Token calls exceeding 700 billion.

At the model level, Baidu officially released the preview version of ERNIE 5.5 Flash. The model has a total parameter scale of 280B and activates only 14B per computation, performing outstandingly in programming, 3D modeling, knowledge work, content creation, and other areas. In terms of application models, the ERNIE Lightning deep thinking model and the ERNIE Yijing micro-expression large model were officially released and simultaneously launched on the Baidu AI Cloud Qianfan platform, opening API (application programming interface) services to enterprise customers and partners.

Promoting agent services across all industries

Currently, Baidu Dazi has been upgraded to a unified entry point connecting all of Baidu's professional agents across its entire lineup, fully opening the Harness engine and agent API. It has already been deployed at enterprises such as China Southern Power Grid, and according to statistics from the AI Insight Database, it became the fastest-growing general-purpose agent in the past quarter. Kuku, jointly launched by Baidu Wenku and Baidu Netdisk, along with its overseas version KOOKO, has achieved deep roots locally while serving the globe, backed by massive professional content, Chinese and English literature, and expert resources, with data, architecture, and models evolving in concert.

The digital human agent Baidu Yijing has launched "Short Comic Drama Studio," covering a one-stop creation pipeline from ideation, script, and storyboarding to the final cut. On Baidu Yijing, creators can localize the plot, characters, and scenes of proven works with one click, helping short comic dramas go global.

Baidu Famou has been upgraded to version 3.0, targeting complex tasks such as production operations, business strategy, and industrial R&D, continuously self-evolving, and taking responsibility for the ultimate business metrics of industrial tasks. Currently, Baidu Famou has industrial customers from 25 major industrial categories.

Strengthening the computing foundation and accelerating global operations

At this year's conference, Baidu AI Cloud released the Tianchi 1024-card super node. At the infrastructure level, Baidu AI Cloud will support a variety of domestic chips, an open super node architecture, and an open-source reinforcement learning framework. Currently, Kunlunxin M100 has been launched on schedule, and M300 will be released next year. Among them, M100 focuses on cost-effectiveness; while M300, as the flagship product, has achieved all-around capability leaps in multiple dimensions such as computing power, storage, and interconnection. Li Yanhong stated that Kunlunxin is the result of Baidu's more than 10 years of sustained investment, and its commercial value is gaining increasingly broad recognition.

In autonomous driving, Apollo Go has been deployed in 28 cities at home and abroad, with autonomous driving mileage exceeding 350 million kilometers, of which fully driverless mileage exceeds 240 million kilometers, and has launched fully driverless commercial operations in Dubai and Abu Dhabi. Apollo Go has also joined hands with the CAR Inc. team, committed to innovating car rental services and creating new car rental models.

Baidu believes that through perceiving changes, rapid experimentation, continuous learning, and scaled replication, "intelligent organizations" will embrace more opportunities in asset-light operations, building new "moats," and creating and discovering new markets. In building "intelligent organizations," Baidu is willing to work with partners from all sectors to re-understand, design, and create organizations, and jointly build the "new factory" of the AI era.

Data source: Baidu

Source

People's Daily