Model release · Baidu
ERNIE 5.5 Flash Preview launches, Qianfan API 50% off for a limited time
- Released
- Review updated
It is the first post-trained model built on the earlier ERNIE 5.5 Flash Base, with 280B total parameters, 14B active, and a 1M-token context. Baidu says it is strong at coding, 3D modeling, knowledge work and content creation, scores competitively on major benchmarks, and decodes quickly. The company also calls this an intermediate post-training checkpoint: training is not finished, some capabilities may still fall short, and the official release will take further improvements.
Availability
The API is live on Baidu Qianfan as ERNIE-5.5-Flash-Preview (endpoint ernie-5.5-flash-preview), with a 1M-token context and 1M max input and output. Qianfan docs say thinking is forced on for this endpoint and cannot be turned off. Usage is 50% off for a limited time: at peak hours, 1 CNY per million input tokens (list 2) and 4 CNY per million output tokens (list 8); off-peak, 0.5 CNY input (list 1) and 2 CNY output (list 4). Cache hits are also 50% off. No end date is given for the promotion. The official release page also links the OpenRouter free endpoint baidu/ernie-5.5-flash-preview:free, served by Qianfan, which is rate-limited and should not be treated as unlimited or permanent. Weights will be open-sourced under Apache 2.0 soon.