Baidu opens ERNIE-5.5-Flash-Base, 280 billion parameters, no fine-tuning

Baidu's ERNIE-5.5-Flash-Base arrives as a deliberately unfinished product: the company released the base weights without fine-tuning or post-training, so the model's usefulness depends on what developers build on top of it. That choice shifts effort to the open-source community, while Baidu says a post-trained version is supposed to follow.

The architecture details explain the design. ERNIE 5.5 uses a mixture-of-experts setup with 280 billion parameters, only 14 billion of which are active per input, and it was pre-trained on 33.5 trillion tokens with a training context of up to 128,000 tokens. Baidu says the model is designed to support a context of up to one million tokens during fine-tuning and inference, a range that matters for long-document and code tasks.

Baidu says that in benchmarks for reasoning, code, and long-context tasks the model competes with similarly sized open foundation models. The weights are on Hugging Face in FP8 and BF16, giving developers a concrete starting point to adapt.