Mathematical Reasoning on AIME 2026 (Accuracy, Drop, TC2b)
97.1AccuracyVibeThinker-3B + CLR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VibeThinker-3B + CLRParams=3B, Post-training optimization=CLR2026.06 | 97.1 | — | — | |
| GLM-5Params=744B2026.06 | 95.8 | — | — | |
| Qwen3.6 PlusParams=N/A2026.06 | 95.3 | — | — | |
| Claude Opus 4.5Params=N/A2026.06 | 95.1 | — | — | |
| VibeThinker-3BParams=3B2026.06 | 94.3 | — | — | |
| DeepSeek V3.2Params=671B2026.06 | 94.2 | — | — | |
| Kimi K2.5Params=1T2026.06 | 93.3 | — | — | |
| GPT-OSS (high)Params=120B2026.06 | 93.2 | — | — | |
| Gemini 3 ProParams=N/A2026.06 | 91.7 | — | — | |
| Qwen3.5-397B-A17BParams=397B2026.06 | 91.3 | — | — | |
| MiniMax M2.7Params=229B2026.06 | 89.8 | — | — | |
| Qwen3-32B (FP16)Model=Qwen3-32B, Precision=FP162026.06 | 70 | — | — | |
| Qwen3-32B (+P + L)Model=Qwen3-32B, Precision=2-bit + FP16 plan + loop rescue2026.06 | 40 | -53.33 | 66.67 | |
| Qwen3-32B (+L)Model=Qwen3-32B, Precision=2-bit + loop rescue2026.06 | 36.67 | -53.33 | 66.67 | |
| Qwen3-32B (2bit)Model=Qwen3-32B, Precision=2-bit2026.06 | 16.67 | -53.33 | 66.67 | |
| Qwen3-32B (+P)Model=Qwen3-32B, Precision=2-bit + FP16 plan2026.06 | 16.67 | -53.33 | 66.67 |