Ring-1T
Open trillion-parameter thinking model (about 50B active); introduces IcePop and C3PO++ for stable RL at that scale and solved IMO 2025 problems at silver level.
ASystem RL framework; AIME 2025 93.4, HMMT 2025 86.72, Codeforces 2088, ARC-AGI-1 55.94 per the paper. A preview came at the end of September 2025. MIT.
- Date
- Friday, 10 October 2025
- Lab
- Ant Group (inclusionAI)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| AIME 2025 | 93.4 | company |
| ARC-AGI-1 | 55.94 | company |
| IMO 2025 | silver-medal level agent framework, self-tested | company |
Date from HF weights upload (2025-10-10 UTC); paper v1 2025-10-21. 'First open trillion-scale thinking model' is Ant's claim; Kimi K2 Thinking (1T, open) followed on 2025-11-06. Benchmarks are company-run.
Sources
This record was checked against its sources on 6 October 2026. How we check