General Reasoning on AIME 2024 (Accuracy)
89.7Accuracy (AIME 2024)TeamTR (Heterogeneous)
Evaluation results
| Method | Links | |
|---|---|---|
| TeamTR (Heterogeneous)Size=1.7B+8B+14B2026.05 | 89.7 | |
| TeamTR (Homogeneous)Size=3×8B2026.05 | 88.1 | |
| Qwen3 (thinking)Size=32B2026.05 | 81.4 | |
| Qwen3-A3BSize=30B2026.05 | 80.4 | |
| o4-miniSize=–2026.05 | 79.6 | |
| QwQSize=32B2026.05 | 79.5 | |
| Debate + JudgeSize=3×8B2026.05 | 71.1 | |
| DeepSeek-R1 DistillSize=70B2026.05 | 70 | |
| Role-playSize=3×8B2026.05 | 69.9 | |
| DebateSize=3×8B2026.05 | 68.7 | |
| DAPOSize=8B2026.05 | 41.3 | |
| GRPOSize=8B2026.05 | 39.1 | |
| GPT-4o-miniSize=–2026.05 | 8.1 |