iFLYTEK unveils xinghuo X2.5 with moe architecture and enhanced capabilities

iFLYTEK has launched the Xinghuo X2.5 large model, marking a significant step in its technological development. The model features an advanced MoE architecture with a parameter scale of 293B-A30B, enabling improvements in code writing and intelligent agent capabilities. It also enhances mathematical logic and semantic understanding. The Xinghuo X2.5 model is fully self-controlled, utilizing China's domestic computing power platform for end-to-end training and inference. It has overcome challenges related to long sequences in training and inference. The model is now available on iFLYTEK's open platform for developers and enterprise users. In addition to the main model, iFLYTEK has released two lightweight edge-side models: Xinghuo X2.5-4B and 1.7B. These models support ultra-long context processing up to 1 million Tokens and can be deployed locally, suitable for various applications such as document processing, code assistance, and data analysis.