Ant Group Unveils Ling-1T: A Trillion-Parameter AI Model

    An abstract digital illustration representing artificial intelligence with interconnected nodes and glowing data streams, symbolizing a trillion-parameter model.

    Ant Group Unveils Ling-1T: A Trillion-Parameter AI Model

    Ant Group Unveils Ling-1T: A Trillion-Parameter AI Model

    By Alex Carter — Alex lives at the intersection of data and narrative, translating complex market trends into actionable insights. With a background in economics, he demystifies the numbers that drive our digital future. | Last updated: October 13, 2025

    Short answer: Ant Group, the fintech arm of Alibaba, has released Ling-1T, a new trillion-parameter large language model (LLM). Announced on October 9, 2025, Ling-1T is designed for complex reasoning and logic, reportedly outperforming rivals like OpenAI and DeepSeek in areas such as code generation and mathematics. Open-sourcing Ling-1T aims to accelerate global AI development and solidify China’s leadership in the rapidly advancing artificial intelligence landscape.

    Ant Group’s Bold AI Leap with Ling-1T

    Ant Group, the prominent Chinese fintech affiliate of Alibaba Group Holding, recently unveiled and open-sourced its groundbreaking trillion-parameter large language model, Ling-1T. This significant announcement on Thursday, October 9, 2025, marks a pivotal moment in the intensely competitive global race for advanced artificial intelligence capabilities. Ling-1T is positioned as a versatile, general-purpose LLM, specifically engineered to excel in intricate reasoning and logic-based tasks across various domains. The launch underscores Ant Group’s commitment to pushing the boundaries of what AI can achieve.

    The model’s technical prowess is considerable, featuring a staggering one trillion total parameters, with approximately 500 billion activated per token during inference. It was rigorously trained on over 20 trillion inference-intensive tokens, a monumental dataset enabling its sophisticated capabilities. Ant Group asserts that Ling-1T surpasses several established open-source and closed-source competitors, including models from DeepSeek, Moonshot AI, and even OpenAI, in crucial performance metrics. These include enhanced code generation, more efficient software development, and advanced competition-level mathematics.

    Performance Benchmarks and Strategic Openness

    In a demonstration of its advanced mathematical reasoning, Ling-1T achieved an impressive 70.42% accuracy on the challenging American Invitational Mathematics Examination (AIME) benchmark. This performance is notably comparable to that of Google’s highly regarded Gemini-2.5-Pro model, highlighting Ling-1T’s sophisticated problem-solving abilities. Furthermore, the model has shown superior scores on platforms like LiveCodeBench, reinforcing its strength in coding-related tasks. These benchmarks position Ling-1T as a leading contender in the global LLM landscape.

    Ling-1T represents Ant Group’s second trillion-parameter model, following the earlier open-source Ring-1T-preview released in September. It is a flagship component of Ant Group’s broader Ling AI model family, also known as BaiLing. This family encompasses three distinct series: the Ling non-thinking models, which include Ling-1T; the Ring thinking models; and the multimodal Ming series. Ant Group’s Chief Technology Officer, He Zhengyu, articulated the company’s philosophy, stating a belief that Artificial General Intelligence (AGI) should serve as a public good. He emphasized the goal of providing

    Leave a Reply

    Your email address will not be published. Required fields are marked *