Baidu Releases ERNIE 5.5 Flash, a 280B Model Close to DeepSeek V4 Flash

Baidu has released ERNIE 5.5 Flash, an efficient sparse mixture-of-experts model with 280 billion total parameters and 14 billion activated per token. It was pretrained from scratch on 33.5 trillion tokens, with training context up to 128K tokens and fine-tuning and inference support for 1M-token context, and its base checkpoint is openly available as a starting point for domain adaptation and further training. It matters because it gives researchers and developers another large open base model at a scale close to DeepSeek V4 Flash, evaluated under a shared protocol against comparable models.

• LinuxDo2026-11-05 06:00