Chinese tech giant Baidu launched its new flagship AI model, ERNIE 5.0, in Beijing on Thursday.
The company unveiled the system at its Baidu World 2025 event. It is built to handle text, images, audio, and video together.
Baidu claims its 2.4 trillion-parameter model beats top rivals like OpenAI’s GPT-5.1 on key business tasks. This includes understanding complex documents and charts.
The release is a strategic move to capture the global enterprise AI market. It offers a powerful, homegrown option as Baidu expands its AI products worldwide.
ERNIE 5.0 Claims Enterprise AI Crown with Document and Chart Supremacy
Announced at its Baidu World 2025 event, the new model represents a significant step in the global AI arms race.
ERNIE 5.0 is a proprietary, natively omni-modal system, designed from the ground up to jointly process and generate content across text, images, audio, and video.
With a reported scale of up to 2.4 trillion parameters, it marks a major leap in complexity for Baidu’s model family.
In a direct challenge to Western AI leaders, Baidu positioned ERNIE 5.0 as a superior tool for enterprise applications.
The company released benchmarks suggesting the model outperforms or matches OpenAI’s latest GPT-5.1 and Google’s Gemini 2.5 Pro in specific, high-value domains.
Baidu claims a distinct advantage in document understanding, citing top scores on tests like OCRBench and DocVQA. It also asserts leadership in structured data reasoning, pointing to its performance on the ChartQA benchmark.
This focus on enterprise-critical tasks signals Baidu’s strategy to compete on practical utility rather than just general conversational ability.
For businesses dealing with automated document processing or financial analysis, these capabilities are core requirements.
The model is now available to enterprise clients through Baidu’s Qianfan cloud platform API and to the public via the ERNIE Bot website.
However, the launch was not without issues. AI evaluator Lisan al Gaib posted a mixed review, stating
“ERNIE 5.0 benchmarks looked insane until I tested it… unfortunately it’s RL braindamaged or they have a serious issue with their chat platform / system prompt.”
Baidu’s developer support account quickly responded,
“Thanks for the feedback! It’s a known bug — certain syntax can consistently trigger it. We’re working on a fix. You can try rephrasing or changing the prompt to avoid it for now,” the company acknowledged, demonstrating emphasis on transparent communication with developers.
A Dual-Track Strategy: Premium APIs and Open-Source Alternatives
ERNIE 5.0 is positioned at the premium end of Baidu’s pricing structure, a deliberate move away from the aggressive price cuts that have defined China’s AI market.
API access is priced at approximately $0.85 per million input tokens and $3.40 per million output tokens, aligning it with other top-tier models rather than budget alternatives.
This suggests Baidu is confident in the model’s high-capability performance for complex, multimodal reasoning tasks.
This premium strategy contrasts sharply with Baidu’s recent actions in the open-source community. Just days before the ERNIE 5.0 launch, Baidu released ERNIE-4.5-VL, a multimodal model under the permissive Apache 2.0 license.
This dual-track approach allows Baidu to court both large enterprises with its flagship proprietary system and the broader developer community with capable, unrestricted open-source tools.
BAidu’s strategy reflects a maturing market where a one-size-fits-all approach is no longer sufficient. By offering both a high-end, fully-supported product and a flexible, open alternative, the company can address different segments of the AI ecosystem.
CEO Robin Li framed this vision by emphasizing the need to make AI a core business function. “When you internalize AI, it becomes a native capability and transforms intelligence from a cost into a source of productivity,” he stated at the event.
Global Ambitions: Baidu Pushes AI Ecosystem Beyond China’s Borders
Alongside the flagship model, the company detailed a significant international push for its entire AI application suite.
Baidu is rolling out several key products to global markets, signaling its ambition to become a worldwide AI infrastructure provider, not just a domestic champion. These initiatives are designed to create a comprehensive ecosystem around its foundational models.
Its no-code application builder, now available globally as MeDo, empowers users without technical expertise to create AI-powered tools.
Baidu also launched Create, a one-stop AI workspace for productivity that has already attracted over 1.2 million users worldwide.
Furthermore, the company’s digital human technology, which saw widespread use in China, has debuted in Brazil with plans for expansion into the U.S. and Southeast Asia.
This ecosystem is powered by a growing user base. Baidu’s general-purpose AI agent, GenFlow, was updated to version 3.0 and now serves more than 20 million users, which Baidu claims makes it the world’s largest agent by user count.
This broad adoption provides crucial data and feedback loops for refining its underlying models.
BAidu’s entire strategy is underpinned by a push for vertical integration. The ERNIE 5.0 launch coincided with Baidu’s announcement of new custom Kunlunxin AI chips, designed to power its models and reduce reliance on foreign hardware.
For Baidu, owning the full stack from silicon to software is the ultimate competitive advantage. As Robin Li explained, “AI agents themselves are the most significant applications, and the speed of technological iteration is the only competitive moat.”
This integrated approach, combining advanced models with a global product suite, positions Baidu as a formidable contender in the next phase of AI deployment.
