Tool-integrated Mathematical Reasoning on AIME 25 (Pass@1, TCN, WTN)
66.4Pass@1Qwen3-14B + PRUNETIR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-14B + PRUNETIRPRUNETIR=true2026.05 | 66.4 | 4.4 | 10.2 | |
| Qwen3-8B + PRUNETIRPRUNETIR=true2026.05 | 60.9 | 4.1 | 10.9 | |
| Qwen3-14BPRUNETIR=false2026.05 | 60.4 | 4 | 10.5 | |
| ReTool-32B + PRUNETIRPRUNETIR=true2026.05 | 56.7 | 9.2 | 5.3 | |
| Qwen3-8BPRUNETIR=false2026.05 | 51.9 | 7.7 | 12.8 | |
| ReTool-32BOfficial Release=true, PRUNETIR=false2026.05 | 49.3 | — | — | |
| SimpleTIR-32BOfficial Release=true2026.05 | 49.2 | — | — | |
| START-32BOfficial Release=true2026.05 | 47.1 | — | — | |
| OpenAI o1-previewOfficial Release=true2026.05 | 37.5 | — | — |