_version_ 1866908469910568960
author Wang, Zihan
Liu, Xinzhang
Yao, Yitong
Wang, Chao
Zhao, Yu
Yang, Zhihao
Deng, Wenmin
Jia, Kaipeng
Peng, Jiaxin
Huang, Yuyao
Xiong, Sishi
Jiang, Zhuo
Yu, Kaidong
Hu, Xiaohui
Yao, Fubei
Fang, Ruiyu
Jiang, Zhuoru
Song, Ruiting
Xie, Qiyi
Xue, Rui
He, Xuewei
Xue, Yanlei
Yuan, Zhu
Zhang, Zhaoxi
Huang, Zilu
Wang, Shiquan
Wang, Xin
Wu, Hanming
Wang, Mingyuan
Zhan, Xufeng
Sun, Yuhan
Xing, Zhaohu
Jiang, Yuhao
Yang, Bingkai
Song, Shuangyong
Li, Yongxiang
He, Zhongjiang
Li, Xuelong
author_facet Wang, Zihan
Liu, Xinzhang
Yao, Yitong
Wang, Chao
Zhao, Yu
Yang, Zhihao
Deng, Wenmin
Jia, Kaipeng
Peng, Jiaxin
Huang, Yuyao
Xiong, Sishi
Jiang, Zhuo
Yu, Kaidong
Hu, Xiaohui
Yao, Fubei
Fang, Ruiyu
Jiang, Zhuoru
Song, Ruiting
Xie, Qiyi
Xue, Rui
He, Xuewei
Xue, Yanlei
Yuan, Zhu
Zhang, Zhaoxi
Huang, Zilu
Wang, Shiquan
Wang, Xin
Wu, Hanming
Wang, Mingyuan
Zhan, Xufeng
Sun, Yuhan
Xing, Zhaohu
Jiang, Yuhao
Yang, Bingkai
Song, Shuangyong
Li, Yongxiang
He, Zhongjiang
Li, Xuelong
contents We introduce the latest series of TeleChat models: \textbf{TeleChat2}, \textbf{TeleChat2.5}, and \textbf{T1}, offering a significant upgrade over their predecessor, TeleChat. Despite minimal changes to the model architecture, the new series achieves substantial performance gains through enhanced training strategies in both pre-training and post-training stages. The series begins with \textbf{TeleChat2}, which undergoes pretraining on 10 trillion high-quality and diverse tokens. This is followed by Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to further enhance its capabilities. \textbf{TeleChat2.5} and \textbf{T1} expand the pipeline by incorporating a continual pretraining phase with domain-specific datasets, combined with reinforcement learning (RL) to improve performance in code generation and mathematical reasoning tasks. The \textbf{T1} variant is designed for complex reasoning, supporting long Chain-of-Thought (CoT) reasoning and demonstrating substantial improvements in mathematics and coding. In contrast, \textbf{TeleChat2.5} prioritizes speed, delivering rapid inference. Both flagship models of \textbf{T1} and \textbf{TeleChat2.5} are dense Transformer-based architectures with 115B parameters, showcasing significant advancements in reasoning and general task performance compared to the original TeleChat. Notably, \textbf{T1-115B} outperform proprietary models such as OpenAI's o1-mini and GPT-4o. We publicly release \textbf{TeleChat2}, \textbf{TeleChat2.5} and \textbf{T1}, including post-trained versions with 35B and 115B parameters, to empower developers and researchers with state-of-the-art language models tailored for diverse applications.
format Preprint
id arxiv_https___arxiv_org_abs_2507_18013
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Technical Report of TeleChat2, TeleChat2.5 and T1
Wang, Zihan
Liu, Xinzhang
Yao, Yitong
Wang, Chao
Zhao, Yu
Yang, Zhihao
Deng, Wenmin
Jia, Kaipeng
Peng, Jiaxin
Huang, Yuyao
Xiong, Sishi
Jiang, Zhuo
Yu, Kaidong
Hu, Xiaohui
Yao, Fubei
Fang, Ruiyu
Jiang, Zhuoru
Song, Ruiting
Xie, Qiyi
Xue, Rui
He, Xuewei
Xue, Yanlei
Yuan, Zhu
Zhang, Zhaoxi
Huang, Zilu
Wang, Shiquan
Wang, Xin
Wu, Hanming
Wang, Mingyuan
Zhan, Xufeng
Sun, Yuhan
Xing, Zhaohu
Jiang, Yuhao
Yang, Bingkai
Song, Shuangyong
Li, Yongxiang
He, Zhongjiang
Li, Xuelong
Computation and Language
I.2.7
We introduce the latest series of TeleChat models: \textbf{TeleChat2}, \textbf{TeleChat2.5}, and \textbf{T1}, offering a significant upgrade over their predecessor, TeleChat. Despite minimal changes to the model architecture, the new series achieves substantial performance gains through enhanced training strategies in both pre-training and post-training stages. The series begins with \textbf{TeleChat2}, which undergoes pretraining on 10 trillion high-quality and diverse tokens. This is followed by Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to further enhance its capabilities. \textbf{TeleChat2.5} and \textbf{T1} expand the pipeline by incorporating a continual pretraining phase with domain-specific datasets, combined with reinforcement learning (RL) to improve performance in code generation and mathematical reasoning tasks. The \textbf{T1} variant is designed for complex reasoning, supporting long Chain-of-Thought (CoT) reasoning and demonstrating substantial improvements in mathematics and coding. In contrast, \textbf{TeleChat2.5} prioritizes speed, delivering rapid inference. Both flagship models of \textbf{T1} and \textbf{TeleChat2.5} are dense Transformer-based architectures with 115B parameters, showcasing significant advancements in reasoning and general task performance compared to the original TeleChat. Notably, \textbf{T1-115B} outperform proprietary models such as OpenAI's o1-mini and GPT-4o. We publicly release \textbf{TeleChat2}, \textbf{TeleChat2.5} and \textbf{T1}, including post-trained versions with 35B and 115B parameters, to empower developers and researchers with state-of-the-art language models tailored for diverse applications.
title Technical Report of TeleChat2, TeleChat2.5 and T1
topic Computation and Language
I.2.7
url https://arxiv.org/abs/2507.18013