List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (Pass@1 Accuracy)

98.6Pass@1 AccuracyNemotron-Cascade-2 30B-A3B
17.4838.5459.680.66Oct 5, 2025Nov 13, 2025Dec 23, 2025Feb 1, 2026Mar 12, 2026Apr 21, 2026May 31, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Nemotron-Cascade-2 30B-A3BTool-Integrated Reasoning (TIR)=true2026.0398.6
Nemotron-Cascade-2 30B-A3B2026.0392.4
Qwen3.5 35B-A3BOfficial/Recommended Settings=true2026.0391.9
Nemotron-3-Super 120B-A12B2026.0390.2
Nemotron-3-Nano 30B-A3B2026.0389.1
SeLaRBackbone=Qwen3-8B2026.0480
SeLaRBackbone=Qwen3-32B2026.0480
CAPSModel=Qwen3-14B, N=16, Temperature=1.02026.0580
V1-InferModel=Qwen3-14B, N=16, Temperature=1.02026.0576.7
V1-InferModel=Qwen3-4B-Instruct-2507, N=16, Temperature=1.02026.0576.7
SwiRBackbone=Qwen3-32B2026.0473.75
CoTBackbone=Qwen3-32B, Decoding Strategy=Greedy2026.0473.33
Mem2EvolveBackbone=GPT-5-Chat2026.0473.33
PointwiseModel=Qwen3-14B, N=16, Temperature=1.02026.0573.3
CAPSModel=Qwen3-4B-Instruct-2507, N=16, Temperature=1.02026.0573.3
CoTBackbone=Qwen3-32B, Decoding Strategy=Sampling2026.0472.08
VanillaModel=Qwen3-14B, N=16, Temperature=1.02026.0570.4
PointwiseModel=Qwen3-4B-Instruct-2507, N=16, Temperature=1.02026.0570
Qwen3-32BStudent Model=-, Teacher Model=Qwen3-32B, Method=Base2026.0568.33
CoTBackbone=Qwen3-8B, Decoding Strategy=Sampling2026.0466.67
Soft ThinkingBackbone=Qwen3-8B2026.0466.67
SwiRBackbone=Qwen3-8B2026.0466.67
AlitaBackbone=GPT-5-Chat2026.0466.67
VanillaModel=Qwen3-4B-Instruct-2507, N=16, Temperature=1.02026.0566.6
Soft ThinkingBackbone=Qwen3-32B2026.0466.25
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=OmniOPD2026.0565.63
Gemini-2.5-flashStudent Model=-, Teacher Model=Gemini-2.5-flash, Method=Base2026.0565.2
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=OmniOPD2026.0564.37
CoTBackbone=Qwen3-8B, Decoding Strategy=Greedy2026.0463.33
AFLOWBackbone=GPT-5-Chat2026.0463.33
SFTStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=SFT2026.0563.13
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OmniOPD2026.0562.92
Qwen3-4B (GRPO)Student Model=Qwen3-4B, Teacher Model=-, Method=GRPO2026.0560.7
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0560.42
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OPD2026.0560.25
GPT-5-Chat (ReAct)Backbone=GPT-5-Chat2026.0460
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=SFT2026.0558.95
CAPSModel=Qwen3-4B-Thinking-2507, N=16, Temperature=1.02026.0556.7
GPT-5-Chat (CoT)Backbone=GPT-5-Chat2026.0456.67
SFTStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=SFT2026.0553.33
V1-InferModel=Qwen3-4B-Thinking-2507, N=16, Temperature=1.02026.0553.3
DSPyBackbone=GPT-5-Chat2026.0450
AgentVerseBackbone=GPT-5-Chat2026.0450
CAPSModel=GPT-OSS-20B, N=16, Temperature=1.02026.0550
PointwiseModel=Qwen3-4B-Thinking-2507, N=16, Temperature=1.02026.0546.7
V1-InferModel=GPT-OSS-20B, N=16, Temperature=1.02026.0546.7
GPT-5-Chat (Direct)Backbone=GPT-5-Chat2026.0446.67
Qwen3-30B-A3B-Inst.Student Model=-, Teacher Model=Qwen3-30B-A3B-Inst., Method=Base2026.0544.79
Qwen3-1.7B (GRPO)Student Model=Qwen3-1.7B, Teacher Model=-, Method=GRPO2026.0544.45
VanillaModel=Qwen3-4B-Thinking-2507, N=16, Temperature=1.02026.0543.7
DyLANBackbone=GPT-5-Chat2026.0443.33
EvoAgentBackbone=GPT-5-Chat2026.0443.33
Qwen3-4BStudent Model=Qwen3-4B, Teacher Model=-, Method=Base2026.0541.88
HEALModels=Qwen2.5-14B-Instruct2026.0340.17
SwarmAgenticBackbone=GPT-5-Chat2026.0440
PointwiseModel=GPT-OSS-20B, N=16, Temperature=1.02026.0540
OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OPD2026.0537.58
SFPOModel=DS-distilled-Qwen-7B2025.1037.5
Qwen3-1.7BStudent Model=Qwen3-1.7B, Teacher Model=-, Method=Base2026.0537.5
VanillaModel=GPT-OSS-20B, N=16, Temperature=1.02026.0537.4
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OPD2026.0536.87
Soft ThinkingBackbone=Qwen3-1.7B2026.0436.67
SeLaRBackbone=Qwen3-1.7B2026.0436.67
AutoAgentsBackbone=GPT-5-Chat2026.0436.67
GRPOModel=DS-distilled-Qwen-7B2025.1035.83
Claude-4.5-haikuStudent Model=-, Teacher Model=Claude-4.5-haiku, Method=Base2026.0534.79
Curriculum SFTModels=Qwen2.5-14B-Instruct2026.0333.33
CoTBackbone=Qwen3-1.7B, Decoding Strategy=Sampling2026.0433.33
OmniOPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0533.33
HEALModels=Qwen3-4B-Base2026.0333.23
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=SFT2026.0531.04
SFPOModel=DS-distilled-Qwen-1.5B2025.1030.83
SFTStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=SFT2026.0530.2
CoTBackbone=Qwen3-1.7B, Decoding Strategy=Greedy2026.0430
EvoMemBackbone=Qwen3-8B, Team size (k)=32026.0530
EVOCHAMBERBackbone=Qwen3-8B, Team size (k)=32026.0530
SFTModels=Qwen2.5-14B-Instruct2026.0329.53
NFPOBase Model=Qwen3-8B-Base, Algorithm=NFPO2026.0528.4
AgentNetBackbone=Qwen3-8B, Team size (k)=32026.0526.7
DPPOBase Model=Qwen3-8B-Base, Algorithm=DPPO2026.0525.6
BaseModel=DS-distilled-Qwen-7B2025.1025
GRPOModel=DS-distilled-Qwen-1.5B2025.1023.33
SwiRBackbone=Qwen3-1.7B2026.0423.33
MemCollabBackbone=Qwen3-8B, Team size (k)=32026.0523.3
CERBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0323.1
GRPOBase Model=Qwen3-8B-Base, Algorithm=GRPO2026.0522.9
Rule-basedBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0322.7
LIMOModels=Qwen3-4B-Base2026.0322.13
General-verifierBackbone=Qwen3-8B-Base, Reward Method=General-verifier, Training Data Domain=General-domain2026.0321.7
Rule-basedBackbone=Qwen3-4B-Base, Training Dataset=mathematical dataset2026.0321.5
CERBackbone=Qwen3-4B-Base, Reward Method=CER, Training Data Domain=General-domain2026.0321.3
Rule-basedBackbone=Qwen3-8B-Base, Reward Method=Rule-based, Training Data Domain=General-domain2026.0321
Rule+CERBackbone=Qwen3-8B-Base, Training Dataset=mathematical dataset2026.0321
GRPOModel=Qwen2.5-Math-7B2025.1020.83
SFPOModel=Qwen2.5-Math-7B2025.1020.83
SFPOModel=Qwen3-4B-Base2025.1020.83
Rule+CERBackbone=Qwen3-4B-Base, Training Dataset=mathematical dataset2026.0320.8
Exact-MatchBackbone=Qwen3-4B-Base, Reward Method=Exact-Match, Training Data Domain=General-domain2026.0320.6
CERBackbone=Qwen3-8B-Base, Reward Method=CER, Training Data Domain=General-domain2026.0320.6
Rule+CERBackbone=Qwen3-8B-Base, Reward Method=Rule+CER, Training Data Domain=General-domain2026.0320.6