Baidu releases ERNIE 5.5 Flash Preview

Baidu's 11 November 2026 release page publishes ERNIE 5.5 Flash Preview, the first post-trained model built on the ERNIE 5.5 Flash Base from earlier. It has 280 billion total parameters, 14 billion active, and a 1 million token context. Baidu says it is aimed at coding, 3D modeling, knowledge work, and content creation, and that it scores competitively on major benchmarks while decoding stays fast. The API is on the Qianfan platform. Baidu says the weights will be open-sourced under Apache 2.0 soon.

Baidu calls the preview an intermediate checkpoint. Post-training is still going, and the page says some capabilities may still fall short. The model is text-only. In one demo on the page, it spent an 11-hour coding task building a third-person game in Unreal Engine 5.8, and because it cannot see the screen it turned playtest problems into tests in code. Other demos on the same page use Blender, FreeCAD, spreadsheets, and a finished video assembled from tools the model calls.

Who it's for: someone calling the Qianfan API for a long coding or document task, or waiting on the Apache 2.0 weights.

Fits: Best for Vibe Coding stays Cursor. Best for Writing stays Claude. A preview checkpoint on Qianfan is a model you call, which leaves the editor and the draft where they are. The coding guide is unchanged. The verdict is unchanged.