Baidu’s ERNIE releases ERNIE 5.5 Flash Preview; model weights to be open-sourced before November 25

IT Home reports that Baidu’s ERNIE large language model has officially launched the ERNIE 5.5 Flash Preview—the first post-training version released after the Base model. With a total parameter count of 280B and 14B activated parameters, it supports a context window of up to 1M tokens and delivers standout performance in programming, 3D modeling, knowledge work, and content creation. The ERNIE 5.5 Flash Preview strikes a balance between intelligence and inference efficiency, achieving industry-competitive scores across multiple benchmark tests while maintaining high decoding speed. Its API is now integrated into Baidu’s Qwen platform, and the model weights are scheduled to be open-sourced by November 25. Baidu emphasizes that this release represents an intermediate outcome of the post-training process: although it already demonstrates strong competitiveness among models of comparable scale, certain capabilities may remain incompletely converged due to training time constraints.