Baidu Releases ERNIE 5.5 Flash Base as 280 Billion Parameter Open Foundation Model

Baidu has released ERNIE 5.5 Flash Base, an open foundation model with 280 billion parameters. The model ships deliberately without fine-tuning or post-training, so developers are expected to decide how to adapt it. A post-trained version is expected to follow.

Model architecture and training

ERNIE 5.5 uses a mixture of experts architecture with 280 billion parameters, 14 billion of which are active per input. It was pre-trained on 33.5 trillion tokens with a training context of up to 128,000 tokens. It is designed to support a context of up to one million tokens during fine-tuning and inference.

Benchmarks and availability

Baidu compared ERNIE 5.5 Flash Base with DeepSeek V4 Flash Base, Hy3 preview Base, and MiMo V2.5 Base using the same evaluation pipeline. The categories were Exam, STEM, Reasoning, Knowledge, Code, Multilingual, and Long Context. Baidu says the model competes with similarly sized open foundation models such as DeepSeek V4 Flash or MiMo V2.5 in reasoning, code, and long-context tasks. The weights are available on Hugging Face in FP8 and BF16.

What developers get

Because the release is a base model, it does not include fine-tuning or post-training. Developers are expected to decide where to take the model, and a post-trained version is supposed to follow.

Sources: The Decoder