List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2024 (Pass@1 Accuracy)

83.33Pass@1 AccuracySeLaR
20.898837.106953.31569.5231Feb 27, 2025May 16, 2025Aug 2, 2025Oct 19, 2025Jan 5, 2026Mar 24, 2026Jun 11, 2026
Updated 4mo ago

Evaluation results

MethodLinks
SeLaRBackbone=Qwen3-8B2026.0483.33
SeLaRBackbone=Qwen3-32B2026.0483.33
SwiRBackbone=Qwen3-32B2026.0482.92
CoTBackbone=Qwen3-32B, Decoding Strategy=Sampling2026.0480.42
CoTBackbone=Qwen3-32B, Decoding Strategy=Greedy2026.0480
Mem2EvolveBackbone=GPT-5-Chat2026.0476.7
DeepSeek-R1-Distill-Qwen-32BModel=DeepSeek-R1-Distill-Qwen-32B, #Bits=W16A162026.0376.67
SliderQuantModel=DeepSeek-R1-Distill-Qwen-32B, #Bits=W4A162026.0376.67
CoTBackbone=Qwen3-8B, Decoding Strategy=Sampling2026.0476.67
Qwen3-32BStudent Model=-, Teacher Model=Qwen3-32B, Method=Base2026.0575.42
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OmniOPD2026.0575
Gemini-2.5-flashStudent Model=-, Teacher Model=Gemini-2.5-flash, Method=Base2026.0574.79
Soft ThinkingBackbone=Qwen3-32B2026.0474.58
DeepSeek-R1-Distill-Qwen-14BModel=DeepSeek-R1-Distill-Qwen-14B, #Bits=W16A162026.0373.33
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=OmniOPD2026.0573.13
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=OmniOPD2026.0572.08
SFTStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=SFT2026.0571.25
SliderQuantModel=DeepSeek-R1-Distill-Qwen-14B, #Bits=W4A162026.0370
CoTBackbone=Qwen3-8B, Decoding Strategy=Greedy2026.0470
Soft ThinkingBackbone=Qwen3-8B2026.0470
AlitaBackbone=GPT-5-Chat2026.0470
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0568.67
Qwen3-4B (GRPO)Student Model=Qwen3-4B, Teacher Model=-, Method=GRPO2026.0568.62
GPT-5-Chat (CoT)Backbone=GPT-5-Chat2026.0466.67
GPT-5-Chat (ReAct)Backbone=GPT-5-Chat2026.0466.67
EvoAgentBackbone=GPT-5-Chat2026.0466.67
AFLOWBackbone=GPT-5-Chat2026.0466.67
DSPyBackbone=GPT-5-Chat2026.0466.67
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OPD2026.0561.45
Qwen3-30B-A3B-Inst.Student Model=-, Teacher Model=Qwen3-30B-A3B-Inst., Method=Base2026.0561.04
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=SFT2026.0560.83
SFTStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=SFT2026.0560.83
SwiRBackbone=Qwen3-8B2026.0460
GPT-5-Chat (Direct)Backbone=GPT-5-Chat2026.0460
AgentVerseBackbone=GPT-5-Chat2026.0460
OmniQuantModel=DeepSeek-R1-Distill-Qwen-32B, #Bits=W4A162026.0356.66
Meta-ReasonerBase Model=DeepSeek-R1-Distill-Qwen-7B2025.0255.5
Claude-4.5-haikuStudent Model=-, Teacher Model=Claude-4.5-haiku, Method=Base2026.0555.2
SFPOModel=DS-distilled-Qwen-7B2025.1054.17
HEALModels=Qwen2.5-14B-Instruct2026.0353.63
SeLaRBackbone=Qwen3-1.7B2026.0453.33
rStar-MathBase Model=Qwen2.5-Math-7B2025.0253.3
Qwen3-4BStudent Model=Qwen3-4B, Teacher Model=-, Method=Base2026.0550.83
GRPOModel=DS-distilled-Qwen-7B2025.1050
OmniQuantModel=DeepSeek-R1-Distill-Qwen-14B, #Bits=W4A162026.0350
TTVSBackbone=Qwen3-8B2026.0450
OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OPD2026.0548.95
Qwen3-1.7BStudent Model=Qwen3-1.7B, Teacher Model=-, Method=Base2026.0548.54
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OPD2026.0548.54
Qwen3-1.7B (GRPO)Student Model=Qwen3-1.7B, Teacher Model=-, Method=GRPO2026.0547.5
OmniOPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0547.08
TTRLBackbone=Qwen3-8B2026.0446.7
CoTBackbone=Qwen3-1.7B, Decoding Strategy=Sampling2026.0446.67
SwiRBackbone=Qwen3-1.7B2026.0446.67
DyLANBackbone=GPT-5-Chat2026.0446.67
SwarmAgenticBackbone=GPT-5-Chat2026.0446.67
rStar-MathBase Model=Phi3-mini-3.8B2025.0243.3
MCTS-RAPBase Model=DeepSeek-R1-Distill-Qwen-7B2025.0240.2
CoTBackbone=Qwen3-1.7B, Decoding Strategy=Greedy2026.0440
AutoAgentsBackbone=GPT-5-Chat2026.0440
Full Fine-TuningRank (r)=N/A, n=642026.0637.03
TTVSBackbone=Qwen3-4B2026.0436.7
Soft ThinkingBackbone=Qwen3-1.7B2026.0436.67
NFPOBase Model=Qwen3-8B-Base, Algorithm=NFPO2026.0536.4
Curriculum SFTModels=Qwen2.5-14B-Instruct2026.0336.27
SFTStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=SFT2026.0535.42
SFPOModel=Qwen2.5-Math-7B2025.1035
DPPOBase Model=Qwen3-8B-Base, Algorithm=DPPO2026.0534.7
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=SFT2026.0534.58
LoRA-αRank (r)=256, n=642026.0634.21
GRPOModel=Qwen2.5-Math-7B2025.1034.2
LoRA (10× LR)Rank (r)=256, n=642026.0633.69
Meta-ReasonerBase Model=GPT-4o2025.0233.3
SFTModels=Qwen2.5-14B-Instruct2026.0333.23
SFPOModel=DS-distilled-Qwen-1.5B2025.1032.5
HEALModels=Qwen3-4B-Base2026.0330.1
GRPOModel=DS-distilled-Qwen-1.5B2025.1030
LoRA-αRank (r)=64, n=642026.0628.69
BaseModel=DS-distilled-Qwen-7B2025.1028.33
LoRA (10× LR)Rank (r)=64, n=642026.0628.29
GRPOBase Model=Qwen3-8B-Base, Algorithm=GRPO2026.0527.4
Qwen3-8BTest-time method=Base2026.0426.9
Rule-basedBackbone=Qwen3-8B-Base, Reward Method=Rule-based, Training Data Domain=General-domain2026.0326.7
General-verifierBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0326.7
SimpleRL-Zero-7BLabeled Data Size=8.9K2026.0426.7
TTRLBackbone=Qwen3-4B2026.0426.7
Meta-ReasonerBase Model=GPT-4o-mini2025.0226.7
LoRARank (r)=256, n=642026.0626.51
Rule+CERBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0326.5
Rule-basedBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0326.3
CERBackbone=Qwen3-8B-Base, Reward Method=CER, Training Data Domain=General-domain2026.0325.8
Exact-MatchBackbone=Qwen3-8B-Base, Reward Method=Exact-Match, Training Data Domain=General-domain2026.0325.4
General-verifierBackbone=Qwen3-8B-Base, Reward Method=General-verifier, Training Data Domain=General-domain2026.0325
DDRLBackbone=Qwen2.5-Math-1.5B2026.0425
CERBackbone=Qwen3-4B-Base, Training Dataset=mathematical dataset2026.0324.8
CERBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0323.8
Rule+CERBackbone=Qwen3-8B-Base, Reward Method=Rule+CER, Training Data Domain=General-domain2026.0323.5
MCTS-RAPBase Model=GPT-4o2025.0223.4
Rule+CERBackbone=Qwen3-4B-Base, Training Dataset=mathematical dataset2026.0323.3
TTVSBackbone=Qwen3-1.7B-Instruct2026.0423.3