TeleAI

China Telecom Artificial Intelligence Technology Co., Ltd., the company behind the TeleAI platform, was established in August 2023 as the artificial intelligence arm of China Telecom Group.^1, 2

In 2023 it released a ten-billion-parameter Xingchen Semantic Model, known as TeleChat, and in January 2024 it published a technical report describing TeleChat models of 3, 7 and 12 billion parameters.^2, 7, 9

The 7B chat model was open-sourced on 10 January 2024, followed by a 12B version on 20 March 2024 and a 1B version on 28 June 2024.⁷

TeleChat2-115B was open-sourced on 20 September 2024, and its repository calls it the first 100-billion-parameter-class model trained entirely on domestic computing power and open-sourced.¹⁰

TeleChat3-105B-A4.7B-Thinking and TeleChat3-36B-Thinking, also trained on domestic computing power, were open-sourced on 23 December 2025.¹¹

In March 2026 the Xingchen AGI Lab was established in Beijing as the research unit for China Telecom's Xingchen foundation models.¹²

Xing4.0-29B-A4B, described as the latest generation of TeleChat and trained on Ascend NPUs, was open-sourced on 17 September 2026 under the Apache 2.0 licence.^3, 8

As of July 2026, Li Xuelong is dean of the China Telecom Institute of Artificial Intelligence (TeleAI) and chief technology officer and chief scientist of China Telecom Group.^5, 6

The platform is operated from Beijing, its API documentation covers OpenAI-compatible endpoints for TeleChat3-438B-A24B and Xing4.0-29B-Intent, and its privacy policy says personal information collected in China is stored in China.^{1, 4, 13, 14}