Baidu World 2025: Why ERNIE 5.0, GenFlow 3.0 and Famou signal a new global AI era

Can Baidu’s AI stack challenge global incumbents with ERNIE 5.0, self-evolving agents, and application-first strategy?

Baidu, Inc. (NASDAQ: BIDU; HKEX: 9888) has taken a significant leap in its global artificial intelligence roadmap with the launch of ERNIE 5.0, an advanced foundation model natively built for multimodal input and output across text, image, audio, and video formats. The announcement was made at Baidu World 2025, the company’s annual technology showcase, alongside a sweeping array of product upgrades, platform launches, and international expansion initiatives.

The Chinese AI and internet giant introduced major enhancements to its flagship productivity tools, AI agents, no-code development platforms, and digital human technologies. Key highlights included the unveiling of GenFlow 3.0, an AI agent now used by over 20 million people globally, the launch of Oreate, a full-suite AI workspace for global markets, and the commercial debut of Famou, a self-evolving agent capable of real-time decision-making in complex domains such as transportation and finance.

Co-founder and Chief Executive Officer Robin Li framed Baidu’s announcements as part of a broader transition toward native AI integration across workflows. He stated that artificial intelligence is moving from being a support tool to becoming a foundational productivity engine. Li emphasized that when AI is embedded directly into enterprise operations, it can transform not only how tasks are executed but also how value is created.

How ERNIE 5.0 redefines foundation model architecture through native multimodality

ERNIE 5.0 was presented as the most advanced evolution of Baidu’s ERNIE model family, and is described as natively omni-modal. Unlike traditional large language models that incorporate multiple modalities via external adapters or alignment techniques, ERNIE 5.0 is architected from the ground up to process, understand, and generate text, image, audio, and video content within a unified framework.

The model is now available through Baidu’s ERNIE Bot for general users and the Qianfan Model-as-a-Service platform for enterprise clients. Key capabilities include advanced instruction following, factual reasoning, agentic planning, and real-time tool use. Robin Li highlighted the rapid pace at which foundation models are evolving, particularly in terms of internal memory depth, problem-solving ability, and their capacity to self-adapt.

According to Baidu’s leadership, AI agents are now becoming the most valuable applications within the ecosystem, and future competitiveness will depend not just on model accuracy or scale, but on the ability to iterate and deploy rapidly across diverse real-world settings.

Why GenFlow 3.0 is becoming the world’s most used general-purpose AI agent

Baidu’s general AI agent GenFlow has undergone a major update with version 3.0, which introduces enhanced multimodal capabilities and persistent memory. The agent supports dynamic interaction across various content types, allowing users to input combinations of text, images, video, and audio while receiving rich, context-aware outputs.

With more than 20 million active users, GenFlow 3.0 now claims the title of the world’s largest general-purpose AI agent by user base. Its utility spans knowledge work, task automation, creative production, and customer interaction. The company positioned GenFlow as a critical component of its application-first AI strategy, enabling developers and enterprises to embed agentic workflows into existing platforms.

Baidu also showcased the integration of GenFlow into broader productivity ecosystems, with plans to introduce APIs and SDKs for customized use across education, customer service, and enterprise verticals.

How Apollo Go is consolidating global leadership in autonomous ride-hailing

Baidu’s Apollo Go robotaxi service has surpassed 17 million cumulative rides, making it the largest fully driverless ride-hailing network in the world. The service operates across 22 cities, with weekly ride volumes exceeding 250,000 and a total of more than 240 million kilometers logged, including over 140 million kilometers in fully autonomous mode.

Robin Li described Apollo Go as a disruptive force in urban mobility. He stated that autonomous vehicles are not only transforming transportation but are also poised to become mobile living spaces. As vehicle costs fall and regulatory frameworks mature, Baidu expects robotaxis to play a central role in future smart city infrastructure.

Investor sentiment around Apollo Go remains cautiously bullish. While monetization pathways are still evolving, analysts believe the platform’s operational data scale and regulatory partnerships provide Baidu with a strategic edge in global autonomous driving markets.

What’s driving Baidu’s AI transformation of its core search engine?

Baidu Search has undergone one of the most aggressive AI-driven redesigns in the global search industry. Approximately 70 percent of top-ranking results are now displayed in rich media formats, including video, images, and interactive panels. This reimagining transforms Baidu Search from a traditional keyword-link service into a dynamic AI-powered discovery engine.

Robin Li explained that Baidu did not simply insert AI summaries into existing formats. Instead, the company rebuilt the interface around content-rich interactions. Baidu has also opened up its AI search architecture via APIs, enabling over 625 ecosystem partners including smartphone manufacturers such as Samsung, Honor, and vivo to incorporate intelligent search functionality into their products.

This architectural openness positions Baidu to become a foundational player in embedded AI search, particularly across mobile and device ecosystems in Asia and beyond.

How Baidu is scaling digital human technology and no-code development tools globally

Baidu’s digital human technology was another focal point at Baidu World 2025. The latest real-time digital human platform allows for contextual interaction, natural emotional expression, and synchronized multi-modal feedback. It has already been widely deployed in China’s livestream commerce sector.

During the recent Double 11 shopping festival, 83 percent of livestreamers adopted Baidu’s digital human tool, resulting in a 119 percent year-over-year increase in active streams and a 91 percent boost in gross merchandise volume. The technology is now being exported, with Brazil as the first test market. Expansion into the United States and Southeast Asia is underway, particularly on e-commerce platforms such as Shopee and Lazada.

Complementing this, Baidu upgraded its no-code app development suite Miaoda to version 2.0, with more than 400,000 applications already built. Its international version, MeDo, was formally launched at the event and is now available for global developers at medo.dev.

What is Oreate and how does it serve Baidu’s international AI productivity goals?

Baidu introduced Oreate, a comprehensive AI workspace platform designed for end-to-end productivity across content types. The tool allows users to create, edit, and publish documents, presentations, videos, podcasts, and images within a single multi-agent system.

Oreate has already attracted over 1.2 million users across global markets, reflecting rising demand for integrated AI tools that bridge creative and operational workflows. The platform aims to compete with international productivity suites by offering built-in multimodal capabilities and native support for agent coordination.

The launch of Oreate also underscores Baidu’s long-term plan to position itself as a full-stack productivity player in the AI economy, extending its reach beyond China’s domestic market.

Why Famou represents a breakthrough in self-evolving AI agents

The most forward-looking announcement at Baidu World 2025 was the launch of Famou, a commercially available self-evolving AI agent. Designed for high-complexity problem environments, Famou can simulate algorithmic logic, abstract domain challenges, and iteratively update solutions based on changing data inputs.

Robin Li stated that Famou can outperform top algorithm engineers in specific contexts and adapt in real time without retraining. Its early use cases include transport routing, energy distribution, financial modeling, and supply chain logistics. Access to Famou is currently limited to invitation-only codes, but Baidu plans a phased rollout to enterprise customers in 2026.

Industry observers view Famou as a major leap toward autonomous AI systems that operate without predefined scripts or developer intervention. If successful at scale, such agents could dramatically reduce infrastructure latency in decision-making environments.

What’s the investor and enterprise outlook after Baidu World 2025?

Institutional response to Baidu World 2025 has been measured but optimistic. Analysts tracking the stock believe Baidu’s strategic focus on interoperable AI ecosystems, rather than stand-alone model competition, provides differentiation in an increasingly saturated AI landscape.

With key products such as GenFlow, MeDo, Oreate, and digital human technology now positioned for international rollout, Baidu is attempting to replicate its domestic AI success abroad. The company’s willingness to expose core infrastructure via APIs and embed into hardware partner ecosystems is also seen as a positive step toward developer trust and adoption.

In the coming quarters, investor focus will likely shift to revenue contribution from ERNIE Bot services, monetization of Apollo Go, and developer traction for tools such as Famou and MeDo. Baidu’s ability to balance model research leadership with practical application-layer deployments will determine whether it can emerge as a global AI platform on par with competitors from the United States and Europe.

Key takeaways from Baidu World 2025 and its global AI expansion push

• Baidu, Inc. launched ERNIE 5.0, a fully multimodal foundation model designed to jointly process text, images, audio, and video from the ground up.
• The model supports advanced capabilities like instruction following, agentic planning, and real-time tool use, and is available through ERNIE Bot and the Qianfan enterprise platform.
• GenFlow 3.0 was unveiled as the world’s most widely used general-purpose AI agent, with over 20 million users and new multimodal and memory enhancements.
• The Apollo Go robotaxi service has completed more than 17 million fully driverless rides across 22 cities, making it the largest global robotaxi network.
• Baidu Search now displays 70 percent of top-ranked results in rich media formats, moving beyond text-link architecture toward an AI-native content discovery experience.
• Baidu launched Oreate, an AI workspace platform for global markets that enables document, image, video, and podcast creation using multi-agent orchestration.
• The company’s no-code platform Miaoda reached 400,000+ apps built, and its global version, MeDo, was launched for developers at medo.dev.
• Baidu’s real-time digital human technology debuted internationally in Brazil and is expanding into Southeast Asia and the United States, with strong adoption in livestream commerce.
• Famou, a self-evolving AI agent capable of simulating expert algorithm logic and adapting autonomously, was introduced for high-complexity use cases in energy, logistics, and finance.
• Baidu positioned its AI ecosystem as an end-to-end stack spanning models, agents, tools, and real-time deployment, with a strong focus on enterprise and developer integration globally.