Baidu introduced two new artificial intelligence (AI) models on Sunday, expanding its AI capabilities. The first model, Ernie 4.5, builds upon the previous generation and serves as a foundation model. The second release, Ernie X1, focuses on reasoning and is noted as the company’s inaugural “deep-thinking” AI model. Baidu asserts that this large language model (LLM) performs comparably to the DeepSeek-R1. In addition to the new models, Baidu announced that access to its AI platform, Ernie Bot, will be free for all users, eliminating any subscription fees for the chatbot.
Baidu Unveils New Ernie AI Models
In a recent press release, Baidu announced the launch of its new AI models. Currently, users can access the Ernie 4.5 model via the chatbot platform, while the Ernie X1 will become available shortly. Unlike many AI firms, Baidu did not publish a research paper or a repository for either of the new models.
Additionally, the company confirmed that Ernie 4.5 will be open-sourced on June 30, allowing developers to access the model’s source code and weights. No such commitment has been made regarding the reasoning model, Ernie X1.
Ernie 4.5 is a multimodal AI model proficient in understanding various formats, including text, images, audio, and videos. This allows users to upload these types of files and pose queries to the AI based on them. Baidu further emphasized that the model’s language proficiency has been improved, enabling it to respond more effectively to conversational input and reasoning-based inquiries. The model is also capable of grasping contextual elements such as memes and satire.
Ernie 4.5 benchmark scores
Photo Credit: Baidu
In terms of performance metrics, Baidu claims that Ernie 4.5 surpasses GPT-4.5 across numerous tests, with pricing set at just one percent of GPT-4.5’s cost. These performance claims are derived from the company’s internal evaluations. Beyond the AI platform, Ernie 4.5 can also be accessed via application programming interfaces on Baidu AI Cloud’s MaaS platform, Quianfan.
Regarding pricing, Baidu stated that input costs for Ernie 4.5 begin at CNY 0.004 (approximately Rs. 0.04) per thousand tokens, while output prices are established at CNY 0.016 (around Rs. 0.19) per thousand tokens.
Turning to Ernie X1, the company reports that this multimodal reasoning model is adept at using various tools and excels in functions such as Chinese knowledge Q&A, creative writing, manuscript composition, conversational dialogue, logical reasoning, and advanced mathematics. Its supported tools encompass capabilities like advanced search, image comprehension, document analysis, image generation, code interpretation, webpage reading, and TreeMind mapping.
Baidu indicated that Ernie X1 was trained using progressive reinforcement learning (RL), employing an end-to-end training model along with a unified multi-dimensional reward system. Like DeepSeek-R1, it features a visible chain of thought (CoT).
Although Ernie X1 is not yet available on the Quianfan platform, once launched, input costs are anticipated to be CNY 0.002 (about Rs. 0.02) and output costs are projected to be CNY 0.008 (roughly Rs. 0.09) per thousand tokens.
