Ant Group Open-Sources Ring-1T-Preview, a Trillion-Parameter Reasoning Model Scoring Higher Than GPT-5

Ant Group has released Ring-1T-Preview, a trillion-parameter natural language reasoning model and the first open-source system of its scale. On the CodeForces coding benchmark, the preview model scored 94.69, above GPT-5, while also achieving 92.6 on AIME 25, close to GPT-5’s 94.6 and ahead of all known open-source models and Gemini 2.5 Pro. Ring-1T-Preview also ranked first among open-source models on LiveCodeBench and ARC-AGI-v1, and demonstrated partial success on IMO25 math competition problems. The model is built on Ant’s Ling 2.0 trillion-parameter foundation, with the full Ring-1T version still in training. [IThome, in Chinese]