Baidu releases ERNIE 5.5 Flash Base with 280 billion parameters as an open-source foundation model
Reported by orcarouter.ai, Unite.AI and The Decoder3 sources ↓
Baidu has released ERNIE 5.5 Flash Base, a 280-billion-parameter mixture-of-experts model pre-trained on 33.5 trillion tokens. It is available without fine-tuning or post-training; Baidu says a post-trained version will follow.
Key points
- The base checkpoint has 14 billion parameters active per token
- Weights are available on Hugging Face in FP8 and BF16 formats
- Training used a context of up to 128,000 tokens; fine-tuning and inference support up to 1 million
- The license excludes the EU, UK and South Korea and sets conditions for commercial use
Sources
Details
Unite.AI reports that Baidu announced the text-only checkpoint on November 5, 2026. It is a pre-training-only model intended as a starting point for domain adaptation, instruction tuning and reinforcement-learning research, rather than a post-trained chat model. Baidu says a post-trained ERNIE 5.5 Flash version will follow.
The model was trained on 33.5 trillion tokens, with training context extended to 128,000 tokens; fine-tuning and inference support up to 1 million tokens. The Decoder reports that Baidu says the model competes with similarly sized open foundation models on reasoning, code and long-context benchmarks. Weights are downloadable from Hugging Face in FP8 and BF16 formats.
Unite.AI reports that the custom license excludes the European Union, United Kingdom and South Korea. Non-commercial use has no model license charge; commercial use tied to annual revenue of up to $50 million requires notice and “Powered by ERNIE” attribution, while use above that threshold requires Baidu’s prior written authorization. The license also restricts using the model, derivatives or outputs to improve unrelated AI or machine-learning systems.
Written from coverage as of
Images
- The Decoder
- orcarouter.ai
- Unite.AI
- orcarouter.ai
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh