In a significant reversal of strategy, Chinese tech giant Baidu has open-sourced its ERNIE 4.5 model family under the Apache 2.0 license, enabling developers worldwide to build AI applications without high costs or vendor lock-in.
The comprehensive release, which became available on June 30, 2025, includes ten distinct model variants. These range from lightweight 0.3 billion parameter dense models to powerful Mixture-of-Experts (MoE) architectures with up to 47 billion active parameters and 424 billion total parameters. Baidu has also released companion developer tools including ERNIEKit for training and fine-tuning, and FastDeploy for efficient deployment across various hardware platforms.
ERNIE 4.5 employs a novel heterogeneous MoE architecture that separates experts for text and vision modalities while sharing knowledge across them. This approach enhances multimodal understanding without compromising performance on text-related tasks. According to Baidu's benchmarks, the ERNIE-4.5-300B-A47B-Base model outperforms DeepSeek-V3-671B-A37B-Base on 22 out of 28 benchmarks, demonstrating strengths in instruction following, world knowledge, visual understanding, and multimodal reasoning.
Industry analysts view this release as a significant development in the global AI landscape. "Every time a major lab open-sources a powerful model, it raises the bar for the entire industry," noted Sean Ren of the University of Southern California. This move puts pressure on closed-source providers like OpenAI and Anthropic to justify their gated APIs and premium pricing models. OpenAI CEO Sam Altman has already acknowledged this shift, indicating plans for an open-source release in the future.
While ERNIE 4.5 doesn't currently lead established models from OpenAI, Google, or DeepSeek in overall performance, its open-source nature and competitive capabilities make it a compelling option for developers. The models are available on multiple platforms including GitHub, Hugging Face, and Baidu AI Studio, with both PaddlePaddle and PyTorch weight formats supported.
Baidu's strategic pivot comes after DeepSeek's open-source models gained international recognition earlier this year, triggering what some call the "DeepSeek moment" in AI development. This trend toward open-source AI is reshaping the industry landscape, potentially democratizing access to advanced AI capabilities and accelerating innovation worldwide.