Baidu Unveils ERNIE 4.5 and ERNIE X1, Challenging AI Rivals
Baidu has entered the next phase of its artificial intelligence development with the launch of its new foundation AI models, ERNIE 4.5 and ERNIE X1. The company is making waves by offering these advanced models to individuals for free via ERNIE Bot, aiming to broaden access to sophisticated AI tools while undercutting the pricing of competitors like OpenAI’s GPT-4.5.
Baidu plans to integrate these models into its product ecosystem, including Baidu Search and the Wenxiaoyan app. The goal is to enhance user experiences across various platforms.
ERNIE 4.5: A Multimodal Powerhouse
ERNIE 4.5, described by Baidu as a “new generation native multimodal foundation model,” boasts improvements in multimodal comprehension through collaborative optimization across numerous modes. It’s designed to enhance language understanding, generation, reasoning, and memory while improving hallucination prevention and coding abilities.
A stand-out feature of ERNIE 4.5 is its capacity to seamlessly integrate and understand diverse content, including text, images, audio, and video. The model is also capable of understanding internet memes and satirical cartoons, showing a high degree of contextual awareness.
Baidu claims that ERNIE 4.5 outperforms GPT-4.5 in various benchmarks, and does so at a fraction of the cost. Currently, Baidu prices its model at only “1% of GPT-4.5.” This breakthrough is attributed to innovations like ‘FlashMask’ dynamic attention masking, heterogeneous multimodal mixture-of-experts, spatiotemporal representation compression, knowledge-centric training data construction, and self-feedback enhanced post-training.
ERNIE X1: Deep-Thinking Reasoning Model
While ERNIE 4.5 emphasizes multimodal capabilities, ERNIE X1 is a deep-thinking reasoning model focused on enhancing understanding, planning, reflection, and evolution. It’s designed as Baidu’s “first multimodal deep-thinking reasoning model capable of tool use,” excelling in applications like Chinese knowledge Q&A, literary creation, and complex calculations.
ERNIE X1 incorporates tool use features such as advanced search, document Q&A, image understanding, AI image generation, and webpage reading. It is fueled by technologies such as the progressive reinforcement learning method, end-to-end training approaches integrating chains of thought and action, and a unified multi-faceted reward system.
Access and Future Plans
For enterprise users and developers, ERNIE 4.5 is accessible through APIs on Baidu AI Cloud’s Qianfan platform, offering competitive pricing. ERNIE X1 will soon be available on the same platform.
Baidu anticipates that 2025 will be a pivotal year for large language model development and aims to continue investing in AI, data centers, and cloud infrastructure to advance its AI capabilities and develop the next generation of models.
