ERNIE 5.5 Flash Preview

What it is

ERNIE 5.5 Flash Preview is a large language model from Baidu. Baidu describes it as the first post-trained model built on ERNIE 5.5 Flash Base ^[1]. Post-training is the further training applied after a base model is pretrained, and it shapes how the model behaves and what it can do.

Origins

Baidu listed ERNIE 5.5 Flash Preview under product updates with the date 11 November 2026 ^[1]. Baidu calls the release an intermediate checkpoint from post-training, so some capabilities may still lag ^[1]. A checkpoint is a saved state of a model during training.

Key features

The model has 280B total parameters, with 14B active parameters ^[1]. Active parameters are the ones used when processing each token. It has a 1M-token context window ^[1], the amount of text it can take into account at once. Baidu says it balances capability with inference efficiency and keeps decoding fast ^[1].

Baidu says the model is aimed at coding, 3D modeling, knowledge work and content creation ^[1]. It also says the model scores competitively across major benchmarks ^[1].

Position in development

As an intermediate checkpoint, the preview shows progress from post-training rather than a finished release. The API is available on Baidu Qianfan ^[1]. Baidu says the weights will be open-sourced under Apache 2.0 soon, and they had not yet been posted ^[1].