BaiduLanguage modelActive
ERNIE 5.5 Flash Base
Summary
ERNIE 5.5 Flash Base is a base model Baidu released on November 5, 2026. It is the checkpoint from the end of pretraining, before any mid- or post-training. Baidu positions it as a starting point for domain adaptation, instruction tuning, and reinforcement learning research. The model handles text only.
It is a sparse mixture-of-experts model with 280 billion total parameters, 14 billion of them activated per token, pretrained on 33.5 trillion tokens. Training ran with up to 131,072 tokens of context, while the model card lists 1,048,576 for fine-tuning and inference. The weights come as FP8 and BF16 variants on Hugging Face and ModelScope under a model-specific license agreement, version 1.0. Local inference examples cover Transformers and vLLM.
All specifications
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- License
- ERNIE 5.5 Flash Base Model License Agreement v1.0. The release specifies this model-specific license.
- Source: License
- Access
- FP8 and BF16 safetensors weights are available for download on Hugging Face and ModelScope. The model card includes local inference examples with Transformers and vLLM.
- Source: Access
Further details
- Checkpoint identity
- ERNIE-5.5-Flash-Base. This is the base checkpoint at the end of pretraining, without subsequent mid- or post-training.
- Source: Checkpoint identity
- Weight repositories
- Baidu publishes FP8 weights at baidu/ERNIE-5.5-Flash-Base and BF16 weights at baidu/ERNIE-5.5-Flash-Base-BF16. These are precision variants of the same base checkpoint.
- Source: Weight repositories
- Architecture and size
- Sparse mixture-of-experts model with 280 billion total parameters and 14 billion parameters activated per token.
- Source: Architecture and size
- Context windows
- The model card lists up to 131,072 tokens for training and 1,048,576 tokens for fine-tuning and inference.
- Source: Context windows
- Pretraining data volume
- Baidu reports 33.5 trillion tokens used in pretraining.
- Source: Pretraining data volume
Frequently asked questions about ERNIE 5.5 Flash Base
How large is the context window of ERNIE 5.5 Flash Base?
1.05 M tokens. That is how much text ERNIE 5.5 Flash Base reads at once, about 1,570 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Can I run ERNIE 5.5 Flash Base locally?
Yes, the weights are open. With 280B parameters it needs a lot of memory, more a server than a laptop.
Stated by Baidu
- Open weights
- yes
- License
- ERNIE 5.5 Flash Base Model License Agreement v1.0. The release specifies this model-specific license.
- Parameters
- 280B
- active per token
- 14B