Mathematical Reasoning on MATH500 (Accuracy, Avg.)
75.5AccuracyCoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoTBackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL Baselines2026.05 | 75.5 | 67.83 | |
| Self-ConsistencyBackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL Baselines2026.05 | 67.6 | 62.42 | |
| TRACERBackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL/RL Baseline Representative2026.05 | 61.2 | 61.85 | |
| Sparse MADBackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL Baselines2026.05 | 47.8 | 55.8 | |
| Single-Agent GRPOBackbone=Qwen2.5-7B-Instruct, Method Category=RL Baselines2026.05 | 47.8 | 46.49 | |
| MADBackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL Baselines2026.05 | 47.6 | 56.11 | |
| Single-Agent GSPOBackbone=Qwen2.5-7B-Instruct, Method Category=RL Baselines2026.05 | 45.8 | 45.63 | |
| Single-Agent GRPOBackbone=Phi-3 Mini 4K Instruct, Method Category=RL Baselines2026.05 | 43.6 | 46.63 | |
| MoABackbone=Qwen2.5-7B-Instruct, Method Category=Non-RL Baselines2026.05 | 43.2 | 53.39 | |
| CoTBackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL Baselines2026.05 | 41.3 | 52.2 | |
| Single-Agent GSPOBackbone=Phi-3 Mini 4K Instruct, Method Category=RL Baselines2026.05 | 41 | 45.72 | |
| TRACERBackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL/RL Baseline Representative2026.05 | 40 | 50.27 | |
| MADBackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL Baselines2026.05 | 37.4 | 51.1 | |
| Sparse MADBackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL Baselines2026.05 | 37.4 | 51.13 | |
| MoABackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL Baselines2026.05 | 36.6 | 47.26 | |
| MAPoRLBackbone=Phi-3 Mini 4K Instruct, Method Category=RL Baselines2026.05 | 36 | 48.93 | |
| Self-ConsistencyBackbone=Phi-3 Mini 4K Instruct, Method Category=Non-RL Baselines2026.05 | 35.8 | 52.32 | |
| MAGRPOBackbone=Phi-3 Mini 4K Instruct, Method Category=RL Baselines2026.05 | 34.2 | 38.52 | |
| MAGRPOBackbone=Qwen2.5-7B-Instruct, Method Category=RL Baselines2026.05 | 32.7 | 34.73 | |
| MAPoRLBackbone=Qwen2.5-7B-Instruct, Method Category=RL Baselines2026.05 | 10 | 38.19 |