These are top 10 stocks traded on the Robinhood UK platform in July
Investing.com -- Baidu (NASDAQ:BIDU) has officially announced that its ERNIE 4.5 series of AI models is now open source, making it available to the global AI community for research and development under the Apache 2.0 license.
The ERNIE 4.5 family includes 10 different variants, ranging from Mixture-of-Experts (MoE) models with 47 billion and 3 billion active parameters to a 0.3 billion dense model. The largest model in the family contains 424 billion total parameters.
Baidu’s move follows similar open-source releases from other Chinese tech giants, including Alibaba (NYSE:BABA)’s Qwen models and DeepSeek’s R1 and V3 models.
The company highlighted three key innovations in the ERNIE 4.5 family: Multimodal Heterogeneous MoE Pre-Training, Scaling-Efficient Infrastructure, and Modality-Specific Post-Training. These innovations allow the models to effectively process both text and visual information.
According to benchmark results provided by Baidu, the ERNIE-4.5-300B-A47B-Base model outperforms DeepSeek-V3-671B-A37B-Base on 22 out of 28 benchmarks across major capability categories. The smaller ERNIE-4.5-21B-A3B-Base model, with approximately 70% of the parameters of Qwen3-30B, outperforms it on several math and reasoning benchmarks.
In multimodal capabilities, the ERNIE-4.5-VL-424B-A47B model demonstrates strong performance across various evaluation benchmarks, particularly in its "thinking mode" for challenging tasks like MathVista and MMMU.
Baidu has released two toolkits to support developers working with these models: ERNIEKit for fine-tuning and alignment, and FastDeploy for efficient model deployment. The models are trained using the PaddlePaddle framework but are also available in PyTorch-compatible formats.
Baidu’s new foray into open source comes at a time when AI competition, both globally and nationally, is at its peak. Baidu ADRs are down 0.8% Monday afternoon.
This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.