Interactive Mathematical Reasoning on GSM8K Insufficient
73.5Success Rate (SR)GRIL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRILModel Scale=Qwen2.5-3B-Instruct2026.04 | 73.5 | 88 | 2.481 | 448 | |
| GRILModel Scale=Qwen3-1.7B2026.04 | 72.8 | 96.5 | 2.348 | 376 | |
| w/ PromptModel Scale=Qwen3-1.7B2026.04 | 66.7 | 88.9 | 2.481 | 744 | |
| w/ SFTModel Scale=Qwen3-1.7B2026.04 | 63.6 | 92.8 | 2.482 | 704 | |
| w/ SFTModel Scale=Qwen2.5-3B-Instruct2026.04 | 62.7 | 90.3 | 2.706 | 575 | |
| GRILModel Scale=Qwen2.5-1.5B-Instruct2026.04 | 61.6 | 90.8 | 2.913 | 479 | |
| w/ PromptModel Scale=Qwen2.5-3B-Instruct2026.04 | 54.9 | 81.5 | 2.919 | 635 | |
| GRILModel Scale=Qwen3-0.6B2026.04 | 52.3 | 84.6 | 2.85 | 1,269 | |
| w/ SFTModel Scale=Qwen3-0.6B2026.04 | 48.4 | 82.4 | 2.971 | 832 | |
| w/ PromptModel Scale=Qwen3-0.6B2026.04 | 44.6 | 79.2 | 2.763 | 1,112 | |
| Base ModelModel Scale=Qwen3-1.7B2026.04 | 41.3 | 52.6 | 3.271 | 1,271 | |
| w/ SFTModel Scale=Qwen2.5-1.5B-Instruct2026.04 | 31.6 | 64.6 | 3.202 | 750 | |
| w/ PromptModel Scale=Qwen2.5-1.5B-Instruct2026.04 | 21 | 45.7 | 3.303 | 579 | |
| Base ModelModel Scale=Qwen2.5-3B-Instruct2026.04 | 20.6 | 28 | 3.665 | 887 | |
| Base ModelModel Scale=Qwen3-0.6B2026.04 | 16 | 24.2 | 3.737 | 1,668 | |
| Base ModelModel Scale=Qwen2.5-1.5B-Instruct2026.04 | 1.8 | 4.6 | 3.828 | 810 |