Baidu has announced the release of two new artificial intelligence (AI) models. The Chinese technology giant introduced Ernie 4.5, a foundation model that builds upon its predecessor, and Ernie X1, a reasoning-focused AI designed for deep thinking. Notably, Baidu has also made its AI-powered chatbot, Ernie Bot, free for users, eliminating any premium charges.
In its official statement, Baidu confirmed that Ernie 4.5 is now accessible via the chatbot platform, while Ernie X1 is expected to be available soon. Unlike many AI firms that accompany new model launches with a research paper and open-source repository, Baidu has not yet published any such documentation. However, it did reveal that the Ernie 4.5 model will be open-sourced on 30 June, providing developers with access to its source code and model weights. There has been no similar announcement regarding Ernie X1.
Ernie 4.5 is a multimodal AI model capable of processing text, images, audio, and videos. Users can input these formats and receive responses to queries related to them. Baidu has emphasised improvements in the model’s language capabilities, enhancing its ability to engage in conversations, respond to reasoning-based queries, and retain memory-based context. Additionally, the model can interpret contextual content such as satire and memes.
In terms of performance, Baidu claims that Ernie 4.5 surpasses OpenAI’s GPT-4.5 across various benchmarks, while being priced at just one per cent of its competitor. These performance metrics are based on Baidu’s internal evaluations. Beyond the chatbot platform, Ernie 4.5 is also accessible via application programming interfaces (APIs) on Baidu AI Cloud’s MaaS platform, Quianfan.
The pricing structure for Ernie 4.5 has also been detailed, with input costs starting at CNY 0.004 (approximately ₹0.04) per thousand tokens and output charges set at CNY 0.016 (around ₹0.019) per thousand tokens.
Meanwhile, Ernie X1 has been developed as a multimodal reasoning model, featuring capabilities such as advanced search, image and document comprehension, code interpretation, webpage analysis, and TreeMind mapping. Baidu has stated that the model excels in Chinese knowledge Q&A, literary composition, manuscript writing, logical reasoning, and complex mathematical problem-solving. It has been trained using progressive reinforcement learning (RL) with an end-to-end training approach and a multi-faceted reward system. Like DeepSeek-R1, it also features a visible chain of thought (CoT) to enhance transparency in decision-making.
Ernie X1 has not yet been integrated into the Quianfan platform, but once launched, its pricing will be set at CNY 0.002 (approximately ₹0.02) for input and CNY 0.008 (roughly ₹0.09) for output per thousand tokens.