Baidu Open-Sources a 280B Model
Baidu released base weights for a sparse MoE model with million-token serving, the third Chinese open release in 48 hours.
Baidu open-sourced ERNIE 5.5 Flash Base on 5 November 2026, a sparse mixture-of-experts model with 280 billion total parameters and 14 billion activated per token, pretrained on 33.5 trillion tokens. Pretraining context reaches 128K, while fine-tuning and inference support up to one million tokens. Baidu claims state-of-the-art results among models of comparable size, citing comparisons against DeepSeek-V4-Flash-Base, Hy3-preview-Base and MiMo-V2.5-Base. Weights are on Hugging Face in FP8 and BF16, released as a base checkpoint without mid- or post-training.
