What changed
Baidu has announced the release of ERNIE 5.5 Flash Base, an open-source foundation model with 280 billion parameters. The model is designed to be used without fine-tuning or post-training, allowing developers to decide how to proceed with its implementation. A post-trained version of the model is expected to be released in the future.
What this means in practice
The ERNIE 5.5 Flash Base model uses a mixture-of-experts architecture with 280 billion total 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, though it is designed to support a context of up to one million tokens during fine-tuning and inference. The model's weights are available on Hugging Face in FP8 and BF16 formats. Baidu claims that the model competes with similarly sized open foundation models like DeepSeek V4 Flash or MiMo V2.5 in reasoning, code, and long-context tasks, based on benchmark results tested with the same evaluation pipeline across multiple categories.
Limitations and availability
The model is currently available as an open-source foundation model, but it is not fine-tuned or post-trained. Developers are expected to take the model and decide how to use it. A post-trained version is planned for release. The model's performance in various benchmarks is reported by Baidu, but independent testing has not been conducted. The availability of the model on Hugging Face provides access to developers, though the exact hardware requirements for running the model are not specified in the announcement.
