Chinese AI beats GPT-5

Baidu says their AI is better than ChatGPT, Gemini. Photo: Baidu .

At the Baidu World 2025 event, Chinese giant Baidu announced its next-generation platform model, a series of AI product upgrades and international expansion plans. Accordingly, ERNIE 5.0 is a proprietary model, capable of processing and creating content in multiple formats such as text, images, audio and video .

The group aims to become a global player in the increasingly competitive AI market. ERNIE 5.0 comes just hours after OpenAI upgraded its flagship platform model from GPT-5 to GPT-5.1, promising to reduce token consumption and improve the interaction experience with more human options.

ERNIE-4.5-VL-28B-A3B-Thinking, the recently released open source model under the enterprise-friendly Apache 2.0 license, users can only access the new chatbot through Baidu's ERNIE Bot website and the Qianfan Cloud Platform API for institutional customers.

The model variant is ERNIE 5.0 Preview 1022, which is optimized for text-heavy tasks, in addition to the general preview version that is capable of balancing multiple data types. In addition, Baidu also introduced a series of important upgrades to the “digital human” platform, no-code tools and general-purpose AI agents.

ERNIE 5.0 marks a shift in how AI is deployed at scale, aiming to expand its AI footprint beyond the Chinese market. “Internalizing AI will become an innate capability and transform AI from a cost to a source of productivity,” said CEO Robin Li.

ERNIE 5.0 shows superiority in understanding graphs and documents. Photo: Baidu.

Test results show that ERNIE 5.0 performs close to or on par with leading Western platform models on a wide range of tasks. ERNIE 5.0 Preview outperforms or is on par with OpenAI's GPT-5-High and Google's Gemini 2.5 Pro on tests of multimodal reasoning, document understanding, and image-based question answering.

At the same time, the model demonstrates strong language and code execution capabilities. The company emphasizes the ability to handle multimodal input-output combinations within the same model, rather than concatenating post-processing data modes, which the team sees as a technical differentiator.

In terms of image tasks, ERNIE 5.0 scored top marks on OCRBench, DocVQA, and ChartQA, three benchmarks that measure document recognition, content understanding, and reasoning with structured data. Baidu said the model outperformed both GPT-5-High and Gemini 2.5 Pro on tests involving documents and charts.

In the imaging domain, ERNIE 5.0 achieved results on par with or exceeding Google Veo3 in categories such as semantic matching and image quality, according to Baidu's internal evaluation. The model's multimodal integration enables the generation and interpretation of visual content with a high level of contextual understanding.

ERNIE 5.0 is positioned at a cost and segment equivalent to other high-end models from Chinese competitors such as Alibaba. It is higher than the previous product such as ERNIE 4.5 Turbo. However, compared to models from the US, the price of the chatbot is still in the mid-range group.

According to Venturebeat, this shows Baidu's strategy of wanting to clearly separate between low-cost models that serve high traffic, and high-performance models for complex tasks and multi-modal reasoning.

Source: https://znews.vn/ai-trung-quoc-danh-bai-gpt-5-post1602692.html