Since the launch of DeepSeek"s R1 model in January, the expectations for new foundation models have substantially increased. Baidu, a prominent Chinese technology firm, has announced the introduction of two new foundation models: ERNIE 4.5 and ERNIE X1.
ERNIE 4.5 is a multimodal foundation model designed to understand text, images, audio, and video. It includes enhancements in language skills, understanding, generation, reasoning, and memory. According to Baidu, ERNIE 4.5 surpasses OpenAI’s GPT-4.5 in various standard benchmarks. A notable aspect of ERNIE 4.5 is its cost, which is just 1% of GPT-4.5"s cost.
Baidu achieved ERNIE 4.5"s performance and cost-efficiency using key technologies 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 is Baidu"s first multimodal deep-thinking reasoning model that is better in Chinese knowledge Q&A, dialogue, logical reasoning, complex calculations, and more. It also supports tool use, including advanced search, Q&A on a given document, image understanding, AI image generation, code interpreting, webpage reading, TreeMind mapping, Baidu academic search, business information search, and franchise information search.
The X1 is powered by key technologies such as the progressive reinforcement learning method, end-to-end training approach integrating chains of thought and action, and a unified multi-faceted reward system. According to Baidu, ERNIE X1 delivers performance on par with DeepSeek R1 at only half the price.
Consumers can access the new ERNIE 4.5 and ERNIE X1 models via ERNIE Bot"s website. Baidu is planning to integrate these new models into Baidu Search, the Wenxiaoyan app, and other services in the coming months. For developers, the new ERNIE 4.5 is now available via APIs on Baidu AI Cloud"s MaaS platform Qianfan, while the ERNIE X1 model will be available soon.
The pricing for ERNIE 4.5 starts at RMB 0.004 per thousand tokens for input, and RMB 0.016 per thousand tokens for output. For ERNIE X1, the input prices begin at RMB 0.002 per thousand tokens, while the output prices are as low as RMB 0.008 per thousand tokens. In terms of US dollars, input and output prices start as low as $0.55 per 1M tokens and $2.2 per 1M tokens.