Baidu releases the pre-training weights of ERNIE 5.5 Flash Base
Baidu's ERNIE release page, headed "Product updates / November 5, 2026", says it is releasing ERNIE-5.5-Flash-Base. It describes a sparse mixture-of-experts language model with 280 billion total parameters, of which 14 billion are activated per token. It says the model was pre-trained from scratch on 33.5 trillion tokens, with a training context of up to 128K, and that fine-tuning and inference support a 1 million token context. It says the weights are the checkpoint at the end of pre-training, before any mid-training or post-training. It says a post-trained ERNIE 5.5 Flash will follow. The page does not give a date for that later model, and it does not state a token price.
Source: Baidu ERNIE (opens in a new tab), 5 Nov 2026
Baidu's ERNIE release page, headed "Product updates / November 5, 2026", says it is releasing ERNIE-5.5-Flash-Base. It describes a sparse mixture-of-experts language model with 280 billion total parameters, of which 14 billion are activated per token. It says the model was pre-trained from scratch on 33.5 trillion tokens, with a training context of up to 128K, and that fine-tuning and inference support a 1 million token context. It says the weights are the checkpoint at the end of pre-training, before any mid-training or post-training. It says a post-trained ERNIE 5.5 Flash will follow. The page does not give a date for that later model, and it does not state a token price.
The page says the weights are on Hugging Face in FP8 and BF16. The example on the page loads baidu/ERNIE-5.5-Flash-Base-BF16. It says use is under the ERNIE 5.5 Flash Base Model License Agreement v1.0. The release page does not set out the commercial terms of that licence.
What it means for business: these are pre-training weights, not a finished chat model. The page says a post-trained version will follow and does not say when. Check the licence file, and any sector rules on Chinese models, before commercial use. This describes the release page. It is not advice on a deployment.