By No Code MBA · Facts checked
- Provider
- Baidu
- Announcement date
- Status at last check
- Generally available
What changed
Baidu describes ERNIE-5.5-Flash-Base as a 280B sparse mixture-of-experts model with 14B activated parameters per token, pretrained from scratch on 33.5T tokens. The released checkpoint is at the end of pretraining, before mid- or post-training, and supports fine-tuning and inference with up to a 1M-token context. These are details from Baidu's model card.
Access and availability
Download the FP8 or BF16 checkpoint from Baidu's official Hugging Face repository after accepting the ERNIE 5.5 Flash Base Model License Agreement. Baidu documents local deployment with vLLM and SGLang; it says no official API service is planned for this base model.
Official access documentationWhat you can build
Treat it as a starting point for adaptation rather than a ready-to-use assistant, and test your own instruction tuning and evaluation before relying on it.
Limitations
This is a base pretraining checkpoint, not an instruction-tuned assistant. Use is governed by Baidu's ERNIE 5.5 Flash Base Model License Agreement, not a standard permissive open-source license. Baidu's benchmark statements are provider-reported.