Baidu Open-Sources a 280B Sparse Model With 14B Active

Baidu released ERNIE-5.5-Flash-Base, a sparse mixture-of-experts model with 280 billion total parameters and roughly 14 billion active per token, pretrained on 33.5 trillion tokens with a 128K default context extendable to 1M through fine-tuning. Baidu frames it as an efficient base for domain adaptation rather than a flagship capability jump. That framing is the commercial pitch: for private deployments the binding cost is fine-tuning and serving, not a benchmark position.