Back to news
launchBaidu2026-11-11

Baidu ERNIE releases 5.5 Flash Preview with 280B parameters and a promise to open weights before November 25

Baidu's ERNIE 5.5 Flash Preview, released November 11, is the first post-training version of the open-weight base model, with 280B total parameters, 14B active parameters, a 1M context window, and weights to be opened before November 25.

Baidu's ERNIE large model released ERNIE 5.5 Flash Preview on November 11, the first post-training version of the base model the company had previously shipped. The model has 280B total parameters and 14B active parameters, supports a 1M context window, and the official description says it performs well in programming, 3D modeling, knowledge work and content creation.

ERNIE 5.5 Flash Preview is said to balance capability with inference efficiency, scoring competitively across a range of benchmarks while keeping decoding speed high. The model API is live on the Qianfan platform, and the weights will be open-sourced before November 25.

Baidu's official note is explicit about the boundaries: the preview is an intermediate version in the post-training process, and although it already achieves competitive results among models of its size, it may still carry capability gaps caused by insufficient training, with post-training still ongoing. The company says it will closely watch real-world performance and user feedback on the preview and refine it in the later official release. That framing makes the build an externally verifiable intermediate product rather than a final form, which makes the commitment to open weights before November 25 the key checkpoint for whether the community can reproduce this round of post-training.

百度文心ERNIE 5.5Flash Preview开源权重