In January 2025, China Telecom Corporation Limited (China Telecom) launched the TeleChat2.5 semantic understanding large model, which was developed on the base of a completely domestically-made 10,000-card cluster and a native deep learning training framework. Its parameter sizes range from hundreds of millions to trillions for a full-scale offering. The official version of the complex reasoning model T1 was also released.
In May, the speech recognition large model was upgraded to understand 60 dialects. Throughout the year, a large-scale visual algorithm repository of over 10,000 visual algorithms has been built based on the visual understanding large model, handling more than 2 billion uses per day.
In October, the first video creation platform powered by the multimodal generation large model and launched by a central state-owned enterprise (SOE) was opened for public use. By then, China Telecom had established China’s first fully homemade AI large model system that is full-modal and full-scale.
Additionally, more than 10 standardized AI products were introduced, including AI-powered cloud disks, cloud computers, and smartphones. China Telecom also founded an “AI+” demonstration base, collaborating with nearly 70 central SOEs to create AI application benchmarks and serving over 20,000 clients in the industry.
