[After 16 Months, Baidu Unveils Massive Model Capabilities Again: 280 Billion Parameter Wenxin 5.5 Flash Open Source] Beating AI News: Baidu has open-sourced the weights of its Wenxin ERNIE-5.5-Flash-Base model. Following the Wenxin 4.5 series in June 2025, Baidu has once again open-sourced a massive model with hundreds of billions of parameters after approximately 16 months. The Wenxin 5.0 and 5.1 models released in between were not open-sourced. Wenxin 5.5 Flash prioritizes inference efficiency, boasting a total of 280 billion parameters, with only 14 billion parameters activated for each generated token. The model was pre-trained from scratch using 33.5 trillion tokens and supports inference context with up to 1 million tokens. Baidu has also redesigned its long text processing mechanism. In the 43-layer network, layer 37 compresses every 128 historical tokens into a single cached record, while the remaining 6 layers retain fine-grained retrieval capabilities for the original content, reducing the GPU memory and computational overhead of long text processing. The official benchmarking compares it with DeepSeek-V4-Flash-Base, Tencent Hy3-preview-Base, and Xiaomi MiMo-V2.5-Base, ranking first in 24 out of 30 benchmark tests. This release is a pure text Base model without in-process or post-processing training, primarily for developers to continue domain fine-tuning, instruction training, and reinforcement learning. Baidu stated that the post-trained Wenxin 5.5 Flash will be released subsequently. The license is stricter than the previous generation. Wenxin 4.5 uses the Apache 2.0 license, while 5.5 Flash uses a custom license, not authorized for use in the EU, UK, and South Korea. For any related product or service to generate over $50 million in revenue over the past 12 consecutive months, separate written authorization from Baidu is required before commercial use. [Original Link]