Skip to content

Baidu opens API access to ERNIE 5.5 Flash Preview

A 280B-parameter, 14B-active mixture-of-experts checkpoint with a 1 million-token context, live now on Qianfan and OpenRouter, with Apache 2.0 weights still to come.

Source: Baidu ERNIE

Baidu has released ERNIE 5.5 Flash Preview, the first post-trained checkpoint built on the ERNIE 5.5 Flash Base model it open-sourced last month. It is a sparse mixture-of-experts model with 280 billion total parameters, 14 billion of them active, and a 1 million-token context window.

The model is text-only: it cannot see images or video, but Baidu’s release notes show it writing FreeCAD scripts to build 3D models and Blender scenes to block out video, alongside stronger results on coding, agentic and knowledge-work tasks. It is live now through Baidu’s Qianfan platform and on OpenRouter. Baidu describes this as “an intermediate checkpoint from post-training”, already competitive with models of its size, with full weights promised under an Apache 2.0 licence “soon”.

Why it matters: another large, permissively licensed mixture-of-experts model with a million-token context is one more option to weigh against Qwen, GLM, DeepSeek and Kimi once self-hosting or fine-tuning is possible, and the API access now lets you test it against your own coding and agentic workloads before the weights land.

The caveat: Baidu is explicit that this is an unfinished checkpoint, not the final release, and there are no independent benchmarks yet to check its claims against.

Share

Related reading

More from The Wire

All briefs