General Reasoning Summary on Aggregate (GSM8K, MATH500, Minerva Math, Olympiad Bench, AIME24, AIME25, GPQA)
79.3AccuracyQwen3-4B-Thinking
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-4B-ThinkingMethod Category=Base Model, Thinking Strategy=Long-CoT2026.01 | 79.3 | 11,566 | — | — | |
| RPAMMethod Category=Data-dependent Merging2026.01 | 75.9 | 5,976 | -4.37 | -48.33 | |
| Ada-R1Method Category=Training-based2026.01 | 75.5 | 6,738 | -4.87 | -41.75 | |
| CoDMethod Category=Prompt-guided2026.01 | 74.7 | 7,329 | -5.86 | -36.64 | |
| Task ArithmeticMethod Category=Data-free Merging2026.01 | 74.6 | 6,440 | -5.98 | -44.32 | |
| TIES MergingMethod Category=Data-free Merging2026.01 | 74 | 6,108 | -6.7 | -47.19 | |
| Average MergingMethod Category=Data-free Merging2026.01 | 73.5 | 6,222 | -7.32 | -46.21 | |
| AIMMethod Category=Data-dependent Merging2026.01 | 73.1 | 6,103 | -7.92 | -47.24 | |
| ACMMethod Category=Data-dependent Merging2026.01 | 71.3 | 6,061 | -10.1 | -47.59 | |
| DARE-LinearMethod Category=Data-free Merging2026.01 | 70.5 | 6,853 | -11.11 | -40.75 | |
| Qwen3-4B-InstructMethod Category=Base Model, Thinking Strategy=Short-CoT2026.01 | 66.4 | 3,884 | -16.37 | -66.42 |