Chinese fintech giant Ant Group has released a new open-source, trillion-parameter large language model (LLM) that boasts stronger maths and coding capabilities than rival models from DeepSeek and OpenAI, escalating the intense competition in artificial intelligence systems.
Advertisement
The new general-purpose Ling-1T model, with 1 trillion parameters, demonstrated “superior complex reasoning ability and overall advantage” against both open-source and closed-source rivals, the company said on Thursday.
It outperformed other major AI models in areas including code generation, software development, competition-level maths and logical reasoning, according to the Hangzhou-based firm.
Ling-1T, for instance, scored higher than DeepSeek-V3.1-Terminus, Moonshot AI’s Kimi-K2-0905, as well as OpenAI’s GPT-5-main on major benchmark platforms including LiveCodeBench and the American Invitational Mathematics Examination (AIME), according to Ant.
Ant is an affiliate of Alibaba Group Holding, the owner of the Post.
Advertisement
Ant also touted that on AIME, Ling-1T achieved an accuracy of 70.42 per cent at an average cost of over 4,000 output tokens per problem, on par with Google’s Gemini-2.5-Pro and surpassing those from DeepSeek, OpenAI and Moonshot.
Ling-1T is Ant’s second LLM with 1 trillion parameters. The company last month released Ring-1T-preview, which it claimed was the world’s first open-source, trillion-parameter thinking model.