Baidu Introduces ERNIE 4.5, Beating DeepSeek and Others

In July 2025, Baidu introduced its most advanced artificial intelligence model, ERNIE 4.5, which it describes as a “native multimodal model.” This groundbreaking system can natively process and understand text, images, audio, and video within a unified framework, enabling the complex integration and processing of different types of data and content generation.

Key Features of ERNIE 4.5

The ERNIE 4.5 model excels in advanced capabilities such as following instructions, using learned information about the world, visual understanding, and versatile reasoning. According to Baidu, it achieves state-of-the-art results across several benchmarks and has demonstrated its superiority over competitors.

Versatile integration is one of the strongest features of the model. ERNIE 4.5 can seamlessly combine and convert information across text, images, audio, and video, enabling contextual and nuanced understanding—for example, interpreting memes or cartoon jokes.

Technical Specifications and Architecture

The model uses a Mixture of Experts (MoE) architecture with up to 47 billion activated parameters, optimizing both performance and computational efficiency. The ERNIE Bot 4.5 series includes 10 different model variants, all of which are open source under the Apache 2.0 license, enabling both academic and commercial use.

ERNIE 4.5 is trained and operated using Baidu's PaddlePaddle deep learning framework, with an impressive 47% model FLOP utilization rate, reflecting its computational efficiency.

Performance and Availability

According to Baidu, ERNIE 4.5 outperforms OpenAI's GPT-4.5 and GPT-4o across multiple industry benchmarks, particularly in multimodal tasks, and is available at a fraction of the cost—approximately 1% of the price of GPT-4. ERNIE 4.5 is even available free of charge for chat use on Baidu's platform.

Reduced hallucinations are another significant advantage—Baidu claims that the model is considerably less prone to generating false or misleading information compared with earlier AI systems.

The Future of the ERNIE Series

Baidu has announced plans for ERNIE 5, with further improvements in multimodal intelligence expected later in 2025. For hands-on testing or more technical details, ERNIE 4.5 is available through Baidu's platform at yiyan.baidu.com.

Its free, open-source nature and low-cost API access have made ERNIE 4.5 widely available to developers, researchers, businesses, and the public, making it a major player in the field of multimodal artificial intelligence.