Archive — CN — technode
Ant Group Open-Sources Ring-1T-Preview, a Trillion-Parameter Reasoning ModelScoring Higher Than GPT-5
Ant Group has introduced Ring-1T-Preview, a groundbreaking trillion-parameter reasoning model now open-sourced. This model surpasses GPT-5 on the CodeForces benchmark with a score of 94.69, performs well on AIME 25 and other benchmarks, and is built on Ant's Ling 2.0 foundation, although the complete version is still under development.
Key points
- Ant Group has launched Ring-1T-Preview, the first open-source trillion-parameter model.
- It scored 94.69 on CodeForces, outperforming GPT-5.
- Achieved a score of 92.6 on AIME 25, close to GPT-5's 94.6.
- Ranked first on LiveCodeBench and ARC-AGI-v1 among open-source models.
- Demonstrated some success on IMO25 math competition problems.
- The model is based on Ant's Ling 2.0 framework, with the complete version still in training.
Why it matters
- The development aligns with 2025 IT trends emphasizing open-source AI and large language models.
- Reflects ongoing competition in AI capabilities between major players like OpenAI, Google, and Ant Group.
- Shows evolution from traditional models to more complex, trillion-parameter systems.
The release of Ring-1T-Preview represents a significant advancement in open-source AI, setting a new benchmark in natural language reasoning and AI competition.