The latest open-weight foundation model from Baidu is designed for faster decoding and stronger performance in tasks such as coding and content creation. The company announced the ERNIE 5.5 Flash Preview, which utilizes a new architecture for inference efficiency, and will make the model weights available by Nov 25.
The system incorporates 280B total parameters—14B of which are active per token—and supports a context window of 1M tokens. The foundation model generates text at speeds exceeding 200 tokens/sec and has posted competitive scores on the AutomationBench and Terminal-Bench benchmarks.
