Baidu recently launched its latest large language models, Ernie 4.5 and Ernie X1, both available for free on the Ernie Bot website. Ernie 4.5, Baidu's first native multi-modal large language model, excels in multi-modal understanding and logical reasoning. Its performance surpasses GPT-4.5 in several benchmark tests, while its API call price is only 1% of GPT-4.5's. This significant cost advantage will undoubtedly attract more developers and businesses.
Ernie 4.5 shows remarkable progress in multi-modal understanding, capable of comprehending and reasoning with graphics, charts, memes, comics, songs, and even movies. In multiple tests, Ernie 4.5 achieved an average score of 79.6, exceeding GPT-4.5's 79.14, demonstrating its strong competitiveness.
Ernie X1, dubbed the "deep thinking model," boasts performance comparable to DeepSeek-R1 and focuses on Chinese knowledge Q&A, literary creation, and logical reasoning. Beyond its "long reasoning chain" advantage, X1 incorporates multi-modal capabilities, understanding and generating images, and utilizing tools to generate code and charts, resulting in richer content. Key technologies like progressive reinforcement learning and a diverse unified reward system enhance its overall reasoning ability and cost-effectiveness.
Both models are now available on the Ernie Bot website for free. Businesses and developers can also access them through Baidu's Intelligent Cloud Qianfan large model platform. Ernie 4.5 is priced at ¥0.004 per thousand tokens for input and ¥0.016 per thousand tokens for output. Ernie X1 offers even more competitive pricing at ¥0.002 per thousand tokens for input and ¥0.008 per thousand tokens for output, highlighting Baidu's strong position in the large language model field.
With continuous technological advancements and increasingly affordable pricing, Baidu's Ernie large language models will bring more convenient and efficient intelligent services to a wider audience, ushering in a new era of AI applications.