Baidu Releases Powerful 280-Billion-Parameter ERNIE AI Model With One-Million-Token Context
The Chinese technology giant has released the open-weight ERNIE-5.5-Flash-Base model, giving developers access to advanced AI technology designed for reasoning, coding and processing extremely large amounts of information.
Baidu has released ERNIE-5.5-Flash-Base, a powerful artificial intelligence model featuring 280 billion parameters, marking another major development in China’s growing AI industry. Announced on November 5, 2026, the model uses a mixture-of-experts architecture that activates approximately 14 billion parameters per token, allowing it to perform complex tasks while reducing computational demands. The company says the model was trained from scratch using 33.5 trillion tokens.
One of the model’s most notable features is its ability to support context lengths of up to one million tokens during fine-tuning and inference. This allows developers to experiment with applications involving enormous documents, extensive code repositories and complex information processing. Baidu also introduced architectural improvements designed to reduce memory requirements and improve efficiency during long-context operations. The model’s weights are available in FP8 and BF16 formats.
According to Baidu, ERNIE-5.5-Flash-Base performed competitively against other leading open-weight models in several benchmark categories, including mathematics, reasoning, coding, multilingual understanding and general knowledge. The company compared it with models such as DeepSeek-V4-Flash-Base, Hy3-preview-Base and MiMo-V2.5-Base. However, these results are based on Baidu’s internal evaluations and have not yet been independently verified. Developers can access the model through platforms supporting frameworks such as vLLM and SGLang.
Despite being described as an open-weight release, the model comes with certain licensing restrictions, including geographical limitations and additional requirements for some commercial users. It is also important to note that this release is a base model without completed instruction tuning or post-training, meaning developers will need additional work before deploying it as a finished conversational assistant. Baidu says a post-trained version, ERNIE 5.5 Flash, will follow. The release strengthens competition among Chinese and international AI developers as the race for more capable and efficient models continues.
