Core Insights - The launch of the national-level AI open-source platform "Huanxin Community" marks a significant step in the construction of China's AI "national team" [1] - ZTE Corporation has contributed 11 core technology achievements, including 6 self-developed large models and 5 industry datasets, enhancing the technological foundation of the "Huanxin Community" [1] - The telecom large model NTele-R1-32B-V1 has shown superior performance in AIME2024 and MATH500 tests, while 3B and 7B multimodal models outperformed larger parameter models in various benchmarks [1] Company Contributions - ZTE has created a complete technology matrix covering "models-data-tools" for the "Huanxin Community" [1] - The TFCE dataset addresses communication scenarios for 4G, 5G, and 6G, filling a gap in the industry [1] - ZTE has achieved compatibility with domestic GPUs, resulting in a 40% improvement in computing efficiency [1]

Core Viewpoint - The "Huanxin Community," a national-level AI open-source platform, was officially launched during the 2025 World Artificial Intelligence Conference, aiming to foster a competitive domestic AI ecosystem through open-source collaboration and innovation [1][2]. Group 1: Platform and Objectives - The "Huanxin Community" focuses on "open-source, collaborative innovation" by providing shared computing resources, model resources, and a development community to lower the barriers to AI technology application [2]. - ZTE Corporation, leveraging its 40 years of communication technology and AI research capabilities, plays a crucial role in the ecosystem's development, committing to deep involvement in model innovation, computing optimization, and practical applications [2][7]. Group 2: Open-source Achievements - ZTE has open-sourced 11 core technological achievements, including 6 self-developed large models and 5 industry-specific datasets, creating a comprehensive technology matrix covering "models, data, and tools" [2][3]. - The NTele-R1-32B-V1 telecom model, trained with only 800 carefully selected samples, achieved a score of 82.5 in the AIME2024 evaluation, surpassing the Qwen3-32B model [3][4]. Group 3: Model Performance - The 3B-Curr-ReFT and 7B-Curr-ReFT models, based on Qwen2.5-VL-Instruct, demonstrated significant performance improvements, with the 3B model achieving an accuracy of 83% in AI2D mathematical reasoning tests, outperforming larger models [4][6]. - The 7B version scored 92.2 in the MathVista evaluation, showing a 33.6 percentage point improvement over the baseline model [4]. Group 4: Industry Impact - The open-sourced datasets cover critical areas such as telecommunications, mathematics, coding, and visual recognition, with the TFCE dataset integrating over 40 years of ZTE's technological expertise [6]. - The collaboration with domestic GPU manufacturers aims to enhance the compatibility of open-source models with domestic chips, achieving a 40% improvement in computing efficiency compared to general solutions [7].

Core Insights - The launch of the "Huanxin Community," a national-level AI open-source platform led by China Mobile and guided by the State-owned Assets Supervision and Administration Commission (SASAC), marks a significant step in the construction of China's AI "national team" [2] - ZTE Corporation has open-sourced 11 core technological achievements, including six self-developed large models and five industry datasets, contributing to the establishment of a domestic AI ecosystem [2][7] - The platform aims to integrate resources from state-owned enterprises, break down technological barriers, and promote inclusive AI development through open-source collaboration [2][7] Group 1: AI Models and Performance - The NTele-R1-32B-V1 telecom model, trained on only 800 carefully selected samples, outperformed industry benchmark models in several authoritative assessments, achieving a score of 82.5 in the AIME2024 evaluation [3][4] - The model demonstrated a 95.2% accuracy rate in the MATH500 test, leading similar models by 1-2 percentage points, showcasing a new paradigm for reducing AI development costs through "small sample efficient training" [3][4] - The open-sourced 7B-Curr-ReFT and 3B-Curr-ReFT models, based on the Qwen2.5-VL-Instruct fine-tuning, exhibited reasoning capabilities comparable to larger models, significantly surpassing existing baselines in multiple public benchmark tests [4][6] Group 2: Datasets and Tools - The five industry datasets cover key areas such as telecommunications, mathematics, code, and visual recognition, with the TFCE dataset being a comprehensive resource for telecommunications AI development [6][7] - The TFCE dataset includes over 1,800 functions and 917 Python problems, providing standardized evaluation scenarios for core telecommunications technologies [6][7] - The "model-data-tool" integrated support system allows developers to quickly build industry solutions by utilizing ZTE's open-source models and accompanying datasets [7] Group 3: Strategic Implications - The collaboration between ZTE and various domestic GPU manufacturers aims to enhance the compatibility of open-source models with domestic chips, improving computational efficiency by 40% compared to general solutions [7] - The active engagement of ZTE in the "Huanxin Community" reflects a deep response from technology enterprises to the national AI strategy, reinforcing the technical foundation of the platform [7] - The synergistic model of "national team + leading enterprises" is expected to propel China's AI industry from a "technology follower" to an "ecosystem leader," injecting strong momentum into the high-quality development of the digital economy [7]

Group 1 - The "Huanxin Community" has officially launched, marking a significant step in the construction of China's AI "national team" [1][2] - The platform aims to integrate resources from state-owned enterprises, break down technical barriers, and promote inclusive AI development [2] - ZTE Corporation has open-sourced 11 core technological achievements, including 6 self-developed large models and 5 industry datasets, contributing to the establishment of a complete technology matrix [2][3] Group 2 - The NTele-R1-32B-V1 telecom model has achieved remarkable results, surpassing industry benchmark models with a score of 82.5 in the AIME2024 evaluation and a 95.2% accuracy in the MATH500 test [3][4] - The open-sourced datasets cover key areas such as telecommunications, mathematics, code, and visual recognition, with the TFCE dataset serving as a comprehensive resource for AI development in the telecommunications sector [7][8] - The collaboration between ZTE and domestic GPU manufacturers has improved computing efficiency by 40% compared to general solutions, enhancing the adaptability of open-source models to domestic chips [8]