Physix FrontierPhysix Frontier · News Briefing Card (QbitAI · Nov 5, 2026)
Baidu open-sources ERNIE 5.5 Flash Base, 280B-parameter MoE
KEY FACTS
- On November 5, Baidu open-sourced its base model ERNIE 5.5 Flash Base.
- The model uses a MoE architecture with 280 billion total parameters and 14 billion parameters activated per token.
- Pretraining used 33.5 trillion tokens, with a pretraining context of up to 128K.
- The fine-tuning and inference stages support a context of up to 1 million tokens.
- Baidu says it achieves SOTA among same-size models on multiple benchmarks covering knowledge, reasoning, and code.
KEY DATA
280 billionTotal parameters
14 billionActivated parameters per token
33.5 trillionPretraining tokens
1 million tokensMax inference context
PHYSIX OBSERVATION
By releasing its flagship base model as open source and emphasizing long context and training efficiency, Baidu shifts the competitive focus from simply stacking parameters to output per unit of compute. For developers, it is a new low-cost option for customizing specialized models, but the quality of its same-size SOTA claims still needs third-party reproduction and verification.
Source: QbitAI report