List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (Acc, Delay, TTFT)

77AccBaseline (Thinking)
-1.343218.995939.33559.6741Oct 1, 2025Oct 29, 2025Nov 26, 2025Dec 24, 2025Jan 21, 2026Feb 18, 2026Mar 19, 2026
Updated 7mo ago

Evaluation results

MethodLinks
Baseline (Thinking)Model=gpt-oss-120b2025.1277326.525,079.68
Baseline (Thinking)Model=Qwen3-235B-A22B (2507)2025.12685,348.225,330.46
AsyncReasoningModel=Qwen3-235B-A22B (2507)2025.1268500.949.06
Baseline (Thinking)Model=Qwen3-30B-A3B (2507)2025.12671,814.091,813.67
AsyncReasoningModel=gpt-oss-120b2025.1266166.7524.63
AsyncReasoningModel=Qwen3-30B-A3B (2507)2025.12625.25.2
Baseline (Thinking)Model=gpt-oss-20b2025.1262523.537,691.55
AsyncReasoningModel=gpt-oss-20b2025.1259147.9624.15
Baseline (Thinking)Model=Qwen3-32B2025.12531,915.21,914.45
Baseline (Interactive)Model=Qwen3-235B-A22B (2507), Protocol=No thinking2025.125320.169.73
Baseline (Interactive)Model=gpt-oss-120b, Protocol=Low budget2025.125191.41,411.24
AsyncReasoningModel=Qwen3-32B2025.1249485.935.41
VISTASeed Condition=Repaired, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0346.67——
VISTASeed Condition=Defective, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0346——
GEPASeed Condition=Defective, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0344——
VISTASeed Condition=Minimal, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0344——
GEPASeed Condition=Minimal, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0342——
No Opt.Seed Condition=Repaired, Base Model=GPT-4.1-mini2026.0340——
No Opt.Seed Condition=Minimal, Base Model=GPT-4.1-mini2026.0340——
GEPASeed Condition=Repaired, Base Model=GPT-4.1-mini, Reflector Model=GPT-4o-mini2026.0339.33——
Baseline (Interactive)Model=Qwen3-30B-A3B (2507), Protocol=No thinking2025.12395.255.25
No Opt.Seed Condition=Defective, Base Model=GPT-4.1-mini2026.0338.67——
Baseline (Interactive)Model=gpt-oss-20b, Protocol=Low budget2025.1236120.091,694.43
Standard TrainingBase Model=Qwen3-14B-Base, Stage=Fully Trained2025.1032.63——
HACPOModel Backbone=Qwen3-8B-Base2026.0332.3——
AlphaRLBase Model=Qwen3-14B-Base, Stage=50%2025.1031.8——
AlphaRLBase Model=Qwen3-14B-Base, Stage=40%2025.1031.25——
AlphaRLBase Model=GLM-4-9B-0414, Stage=50%2025.1030——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=Fully Trained2025.1029.75——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=50%2025.1028.89——
HACPOModel Backbone=Qwen3-4B-Base2026.0327.5——
AlphaRLBase Model=GLM-4-9B-0414, Stage=40%2025.1027.25——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=Fully Trained2025.1024.17——
AlphaRLBase Model=Qwen3-14B-Base, Stage=30%2025.1023.89——
AlphaRLBase Model=Qwen3-8B-Base, Stage=40%2025.1023.75——
AlphaRLBase Model=Qwen3-8B-Base, Stage=50%2025.1023.33——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=50%2025.1022.67——
HACPOModel Backbone=Qwen3-1.7B-Base2026.0322——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=40%2025.1021.5——
AlphaRLBase Model=Qwen3-14B-Base, Stage=20%2025.1021.11——
TTSRBackbone Model=Qwen3-4B-Base2026.0220.1——
Baseline (Interactive)Model=Qwen3-32B, Protocol=No thinking2025.12201.71.13
AlphaRLBase Model=Qwen3-8B-Base, Stage=30%2025.1020——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=50%2025.1020——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=40%2025.1020——
AlphaRLBase Model=GLM-4-9B-0414, Stage=30%2025.1019.33——
TTSRBackbone Model=Qwen3-8B-Base2026.0219.1——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=30%2025.1018.33——
AlphaRLBase Model=GLM-4-9B-0414, Stage=20%2025.1017——
TTRLBackbone Model=Qwen3-8B-Base2026.0215.7——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=20%2025.1015.33——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=30%2025.1015.06——
AlphaRLBase Model=Qwen3-14B-Base, Stage=10%2025.1015.05——
R-ZeroBackbone Model=Qwen3-8B-Base2026.0213.4——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=20%2025.1013.33——
AlphaRLBase Model=Qwen3-8B-Base, Stage=20%2025.1012.5——
TTSRBackbone Model=OctoThinker-8B-Hybrid-Base2026.0212.4——
AlphaRLBase Model=Qwen3-8B-Base, Stage=10%2025.1011.67——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=30%2025.1011.67——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=40%2025.1011.67——
AlphaRLBase Model=GLM-4-9B-0414, Stage=10%2025.1011.67——
AlphaRLBase Model=Qwen3-14B-Base, Stage=5%2025.1011.11——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=10%2025.1010.17——
Base ModelBackbone Model=Qwen3-8B-Base2026.029.8——
TTRLBackbone Model=Qwen3-4B-Base2026.029.7——
R-ZeroBackbone Model=Qwen3-4B-Base2026.029.1——
Standard TrainingBase Model=Qwen3-14B-Base, Stage=5%2025.108.33——
TTRLBackbone Model=OctoThinker-8B-Hybrid-Base2026.028.2——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=10%2025.107.5——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=20%2025.107.5——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=10%2025.107.5——
R-ZeroBackbone Model=OctoThinker-8B-Hybrid-Base2026.026.8——
HACPOModel Backbone=Llama3.2-3B-Instruct2026.036.7——
AlphaRLBase Model=Qwen3-8B-Base, Stage=5%2025.106.67——
Base ModelBackbone Model=Qwen3-4B-Base2026.025.8——
Standard TrainingBase Model=Qwen3-8B-Base, Stage=5%2025.105.5——
Base ModelBackbone Model=OctoThinker-8B-Hybrid-Base2026.023.9——
AlphaRLBase Model=GLM-4-9B-0414, Stage=5%2025.103.33——
HACPOModel Backbone=Llama3.2-1B-Instruct, Experimental Context=Heterogeneous Setup2026.033.3——
HACPOModel Backbone=Llama3.2-1B-Instruct2026.032.2——
Standard TrainingBase Model=GLM-4-9B-0414, Stage=5%2025.101.67——