RELEASE · MODELS · #3163
Baidu open-sources ERNIE-5.5-Flash-Base, a 280B-parameter MoE foundation model
Baidu released ERNIE-5.5-Flash-Base, an open-source foundation model using a mixture-of-experts architecture with 280 billion parameters (14 billion active per input). The base model, pre-trained on 33.5 trillion tokens with a 128k training context and designed to support up to 1M-token context in fine-tuning/inference, is available on Hugging Face in FP8 and BF16; Baidu says it compares with similarly sized open models such as DeepSeek V4 Flash and MiMo V2.5, and a post-trained version is planned.
KEY POINTS
- Baidu released ERNIE-5.5-Flash-Base, an open-source foundation model using a mixture-of-experts architecture with 280 billion parameters (14 billion active per input).
- The base model, pre-trained on 33.5 trillion tokens with a 128k training context and designed to support up to 1M-token context in fine-tuning/inference, is available on Hugging Face in FP8 and BF16; Baidu says it compares with similarly sized open models such as DeepSeek V4 Flash and MiMo V2.5, and a post-trained version is planned.
- An open 280B-parameter MoE base model available on Hugging Face accelerates developer and research work on large-scale models, long-context use cases, and fine-tuning strategies.
WHY IT MATTERS
An open 280B-parameter MoE base model available on Hugging Face accelerates developer and research work on large-scale models, long-context use cases, and fine-tuning strategies.