Ernie Flash Preview Cuts the Long-Document Price
A 280-billion-parameter model with a million-token window, sold at half price until demand finds it.
Baidu’s ERNIE 5.5 Flash Preview carries 280 billion total parameters with 14 billion activated per token and a one-million-token context, generating at more than 200 tokens per second. Qianfan is pricing it at half rate for a limited period, with input at 2 yuan per million tokens during busy hours and 1 yuan off-peak, and output at 8 and 4 yuan respectively. Weights are scheduled for open release on 25 November, and the model is already listed on Arena. The combination of long context and speed pushes the unit cost of long-document processing and long-horizon agent work to a new low, which matters most for teams generating content or creative at volume.