Tele-FLM (also known as FLM-2) is a 52-billion parameter open-source multilingual large language model with a stable and efficient pre-training paradigm and enhanced fact-checking capabilities. Based on a decoder-only transformer architecture, it has been trained on approximately 2 trillion tokens. Tele-FLM exhibits superior performance compared to models of similar size, sometimes even surpassing larger ones. Besides sharing the model weights, we also provide core design, engineering practices, and training details, hoping they will benefit both the academic and industrial communities.