Mathematical Reasoning on AIME25 (Accuracy, C_mem, C_time)
69.2AccuracyStepFlow
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| StepFlowBackbone=GPT-OSS-20B medium2026.04 | 69.2 | — | — | |
| DeepConfBackbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 68.9 | 0.75 | 3.57 | |
| Std-PT.Backbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 68.67 | 1 | 1 | |
| T2Backbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 68.28 | 0.26 | 0.5 | |
| ESCBackbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 68.12 | 0.41 | 2.19 | |
| ACBackbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 67.79 | 0.37 | 2.75 | |
| StepFlowBackbone=DeepSeek-R1-Distill-Qwen 32B2026.04 | 66.7 | — | — | |
| DSCBackbone model=Deepseek-R1-Distill-Qwen-7B2026.01 | 65.3 | 0.32 | 2 | |
| GPT-OSS-20B mediumInference Protocol=Baseline2026.04 | 62 | — | — | |
| StepFlowBackbone=DeepSeek-R1-Distill-Qwen 14B2026.04 | 57.7 | — | — | |
| DeepSeek-R1-Distill-Qwen 32BInference Protocol=Baseline2026.04 | 54.9 | — | — | |
| T2Backbone model=Qwen3-4B2026.01 | 54.15 | 0.17 | 0.56 | |
| Std-PT.Backbone model=Qwen3-4B2026.01 | 54 | 1 | 1 | |
| ESCBackbone model=Qwen3-4B2026.01 | 53.86 | 0.28 | 2.25 | |
| ACBackbone model=Qwen3-4B2026.01 | 53.81 | 0.26 | 2.41 | |
| DeepConfBackbone model=Qwen3-4B2026.01 | 52.97 | 0.24 | 3.76 | |
| DSCBackbone model=Qwen3-4B2026.01 | 50.23 | 0.22 | 2.78 | |
| DeepSeek-R1-Distill-Qwen 14BInference Protocol=Baseline2026.04 | 50.2 | — | — | |
| StepFlowBackbone=QwQ-32B-Preview2026.04 | 48 | — | — | |
| StepFlowBackbone=GPT-OSS-20B low2026.04 | 46.5 | — | — | |
| StepFlowBackbone=DeepSeek-R1-Distill-Qwen 7B2026.04 | 43.8 | — | — | |
| QwQ-32B-PreviewInference Protocol=Baseline2026.04 | 40 | — | — | |
| Budget Forcing (S1)Backbone=DeepSeek-R1-Distill-Qwen 7B2026.04 | 39.5 | — | — | |
| DeepSeek-R1-Distill-Qwen 7BInference Protocol=Baseline2026.04 | 39.2 | — | — | |
| GPT-OSS-20B lowInference Protocol=Baseline2026.04 | 39.1 | — | — | |
| T2Backbone model=Qwen2.5-7B2026.01 | 10.32 | 0.35 | 0.68 | |
| DeepConfBackbone model=Qwen2.5-7B2026.01 | 10.25 | 0.78 | 1.88 | |
| Std-PT.Backbone model=Qwen2.5-7B2026.01 | 10.06 | 1 | 1 | |
| ESCBackbone model=Qwen2.5-7B2026.01 | 10.02 | 0.33 | 1.45 | |
| ACBackbone model=Qwen2.5-7B2026.01 | 9.94 | 0.66 | 2.83 | |
| DSCBackbone model=Qwen2.5-7B2026.01 | 8.89 | 0.28 | 0.97 | |
| Std-PT.Backbone model=Llama3.1-8B2026.01 | 6.57 | 1 | 1 | |
| T2Backbone model=Llama3.1-8B2026.01 | 6.38 | 0.17 | 0.52 | |
| ACBackbone model=Llama3.1-8B2026.01 | 6.08 | 0.41 | 3.5 | |
| ESCBackbone model=Llama3.1-8B2026.01 | 5.99 | 0.21 | 2.44 | |
| DeepConfBackbone model=Llama3.1-8B2026.01 | 4.87 | 0.72 | 5.22 | |
| DSCBackbone model=Llama3.1-8B2026.01 | 4.35 | 0.14 | 0.98 |