Baidu Releases Open-Source Base Model ERNIE 5.5 Flash Base
Understand the story
On November 6, 2026, The Decoder reported that Baidu released its open-source base model ERNIE-5.5-Flash-Base. The model is deliberately not fine-tuned or post-trained, leaving developers to decide its use, with post-trained versions to be released later. Technical details: the model uses a MoE (Mixture of Experts) architecture with 280 billion (280B) total parameters and 14 billion (14B) activated per input; the pretraining data scale is 33.5 trillion tokens; the training context length is up to 128,000 tokens; and in its architecture design, fine-tuning and inference can support context lengths of up to 1 million tokens. Currently this is the only report on the event, with no further developments or conflicting information.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderBaidu releases ERNIE 5.5 Flash Base with 280 billion parameters as an open-source foundation model
Baidu released the open-source foundation model ERNIE-5.5-Flash-Base, deliberately shipped without fine-tuning or post-training so developers decide its use, with post-trained versions to follow. The model uses a MoE architecture with 280B total parameters and 14B activated per input, pretrained on 33.5 trillion tokens with a training context of up to 128,000 tokens, and is designed to support contexts of up to 1 million tokens during fine-tuning and inference.
Story heat trend
Not enough continuous observations to show a trend yet.