Baidu, a major AI and internet company, has unveiled its latest foundation models: ERNIE 4.5, a native multimodal model, and ERNIE X1, a deep-thinking reasoning model. The company noted that this launch highlights its focus on advancing next-generation AI technologies.
Baidu said the release of ERNIE 4.5 and ERNIE X1 marks a “significant milestone in pushing the boundaries of multimodal and reasoning models,” providing “advanced capabilities at a more accessible price point.”
These models will integrate into Baidu’s ecosystem, including Baidu Search and the Wenxiaoyan app, to improve experiences for a wide range of users.
ERNIE 4.5: Multimodal Excellence
ERNIE 4.5, built solely by Baidu, is a next-gen multimodal model. It processes text, images, audio, and video, showing strong comprehension skills. Baidu reported enhancements in understanding, generation, reasoning, and memory, with better hallucination prevention, logical reasoning, and coding abilities.
The model grasps internet memes and satirical cartoons effortlessly, thanks to its contextual awareness. Baidu added that ERNIE 4.5 outperforms GPT-4.5 in multiple benchmarks while costing just 1% of GPT-4.5’s price.
Key technologies driving this include “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: Reasoning with Tools
ERNIE X1, Baidu’s first multimodal deep-thinking reasoning model, excels in understanding, planning, reflection, and evolution. It handles Chinese Q&A, literary creation, manuscript writing, dialogue, logical reasoning, and complex calculations.
Baidu highlighted its tool-use capabilities, supporting advanced search, document Q&A, image understanding, AI image generation, code interpretation, webpage reading, TreeMind mapping, Baidu academic search, business information search, and franchise information search.
Its strengths come from the Progressive Reinforcement Learning Method, End-to-End Training Integrating Chains of Thought and Action, and a Unified Multi-Faceted Reward System, Baidu explained.
Pricing and Availability
- For enterprises and developers, ERNIE 4.5 is available now via APIs on Baidu AI Cloud’s Qianfan platform, with input pricing at RMB 0.004 per 1,000 tokens and output at RMB 0.016 per 1,000 tokens.
- ERNIE X1, arriving soon, will cost RMB 0.002 per 1,000 input tokens and RMB 0.008 per 1,000 output tokens.
Baidu also announced that ERNIE Bot is now free for individual users ahead of its planned April 1 rollout, a shift noted earlier this year. Both models are accessible at no cost on the ERNIE Bot website.
Future Vision
Baidu expects 2025 to be a key year for AI progress. The company stated,
With the launch of ERNIE 4.5 and ERNIE X1, Baidu will continue to invest in artificial intelligence, data centers, and cloud infrastructure to advance our AI capabilities and develop smarter and more powerful next-generation models.