TeleChat-7B

TeleChat-7B is the open-sourced 7-billion-parameter version of China Telecom's XingChen semantic large model, developed by China Telecom Artificial Intelligence Technology Co. It was pretrained on 1.5 trillion Chinese and English tokens, and the release included a 1TB cleaned dataset plus a technical report describing the data cleaning workflow and multi-stage supervised fine-tuning.

The chat-tuned 7B variant was open-sourced in January 2024, followed by a 12B version. It represents one of the earliest open large-model releases from a Chinese state-owned telecom operator, aimed at giving developers a domestic bilingual foundation to build on. Check the model card for current license terms before commercial use.