List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 25 (Pass@1 Accuracy)

89.1Pass@1 AccuracyNemotron 3 Nano 30B-A3B
10.299230.757151.21571.6729Jul 31, 2025Sep 18, 2025Nov 6, 2025Dec 25, 2025Feb 12, 2026Apr 2, 2026May 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Nemotron 3 Nano 30B-A3BOpen-Source=✓, Size=30B-A3B, no tools=true2026.0489.1
Nemotron 3 Nano OmniOpen-Source=✓, Size=30B-A3B, no tools=true2026.0482.1
Qwen3-OmniOpen-Source=✓, Size=30B-A3B, no tools=true2026.0473.7
ROPDThinking Mode=Thinking2026.0568.75
FP16Model=Qwen3-8B, Quantization Bits=FP16, Speedup=1.0×2026.0567.66
GPT-5.2-chat (teacher)2026.0567.08
OVDThinking Mode=Thinking2026.0565.83
T-JudgeThinking Mode=Thinking2026.0565.48
KnowRL-Nemotron-1.5BHint Setting=CSS, Evaluation Protocol=mean@322026.0465.21
KnowRL-Nemotron-1.5BHint Setting=CBRS, Evaluation Protocol=mean@322026.0465
QuestAHint Setting=CSS, Evaluation Protocol=mean@322026.0464.99
KnowRL-Nemotron-1.5BHint Setting=w/o KP, Evaluation Protocol=mean@322026.0464.69
FP16Model=Qwen3-4B, Quantization Bits=FP16, Speedup=1.0×2026.0563.75
JustRLHint Setting=w/o KP, Evaluation Protocol=mean@322026.0462.92
JustRLHint Setting=CBRS, Evaluation Protocol=mean@322026.0462.36
QuestAHint Setting=w/o KP, Evaluation Protocol=mean@322026.0462.08
QuestAHint Setting=CBRS, Evaluation Protocol=mean@322026.0462
JustRLHint Setting=CSS, Evaluation Protocol=mean@322026.0461.43
LAQuantModel=Qwen3-8B, Quantization Bits=3-bit, Speedup=5.15×2026.0561.3
GADThinking Mode=Thinking2026.0561.28
Qwen3-30B-A3B (teacher)Access=–2026.0561.25
Qwen3-4B (student)Thinking Mode=Thinking2026.0559.58
ROPDThinking Mode=Non-Thinking2026.0558.75
ParoQ++Model=Qwen3-8B, Quantization Bits=3-bit, Speedup=4.63×2026.0558.39
T-JudgeThinking Mode=Non-Thinking2026.0556.64
ROPDAccess=text2026.0555.93
OVDThinking Mode=Non-Thinking2026.0555.71
LAQuantModel=Qwen3-4B, Quantization Bits=3-bit, Speedup=3.42×2026.0554.69
ESampModel=GPT-OSS-20B2026.0454.2
Min-pModel=GPT-OSS-20B2026.0453.8
FIREModel=GPT-OSS-20B2026.0453.5
OverRIDEModel=GPT-OSS-20B2026.0453.3
ParoQModel=Qwen3-8B, Quantization Bits=3-bit, Speedup=4.63×2026.0552.92
ParoQ++Model=Qwen3-4B, Quantization Bits=3-bit, Speedup=3.01×2026.0552.76
Nemotron-1.5BHint Setting=CSS, Evaluation Protocol=mean@322026.0450.1
Nemotron-1.5BHint Setting=CBRS, Evaluation Protocol=mean@322026.0449
VanillaModel=GPT-OSS-20B2026.0449
Nemotron-1.5BHint Setting=w/o KP, Evaluation Protocol=mean@322026.0448.33
R-QATModel=Qwen3-8B, Quantization Bits=3-bit, Speedup=5.15×2026.0543.91
FIREModel=Qwen3-8B2026.0442.4
ExOPDAccess=logit2026.0541.25
GPTQModel=Qwen3-8B, Quantization Bits=3-bit, Speedup=5.15×2026.0540.68
ParoQModel=Qwen3-4B, Quantization Bits=3-bit, Speedup=3.01×2026.0539.58
ESampModel=Qwen3-8B2026.0439.1
LOPDAccess=logit2026.0538.75
Min-pModel=Qwen3-8B2026.0438.6
VanillaModel=Qwen3-8B2026.0438
OverRIDEModel=Qwen3-8B2026.0437.8
FP16Model=Qwen3-1.7B, Quantization Bits=FP16, Speedup=1.0×2026.0535.94
R-QATModel=Qwen3-4B, Quantization Bits=3-bit, Speedup=3.42×2026.0535.31
ASLEC-CASLBackbone=Llama3-3B2026.0433.33
AMS-ExpectedKV Cache Length (T_keep)=10242026.0533.33
CASPOBase Model=Qwen3-8B-Base, RM Data=02026.0533.3
FP16Model=R1-Distill-Llama-8B, Quantization Bits=FP16, Speedup=1.0×2026.0531.2
EEPOBase Model=Qwen3-14B-Base2025.1030
Full KVKV Cache Length (T_keep)=Full2026.0530
LAQuantModel=Qwen3-1.7B, Quantization Bits=3-bit, Speedup=2.29×2026.0529.33
LAQuantModel=R1-Distill-Llama-8B, Quantization Bits=3-bit, Speedup=5.56×2026.0528.54
ParoQ++Model=Qwen3-1.7B, Quantization Bits=3-bit, Speedup=2.05×2026.0528.23
R-QATModel=R1-Distill-Llama-8B, Quantization Bits=3-bit, Speedup=5.56×2026.0527.5
SCRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-Math-7B2026.0326.9
Re-ScheduleBackbone=Qwen3-4B-Base2025.1026.9
GRPOBase Model=Qwen3-14B-Base2025.1026.7
SatoriBase Model=Qwen3-8B-Base, RM Data=240K2026.0526.7
SA + GRPOTraining Paradigm=Single Agent2026.0526.67
MetaAgent-X RLTraining Paradigm=RL-based Auto MAS2026.0526.67
AdaKV-ExpE2KV Cache Length (T_keep)=10242026.0526.67
GRACEBackbone=Llama3-3B2026.0426.66
RL-PLUSBase Model=Qwen2.5-Math-7B, Training Method=RL-PLUS2025.0725.9
ParoQ++Model=R1-Distill-Llama-8B, Quantization Bits=3-bit, Speedup=4.90×2026.0525.57
MaASTraining Paradigm=RL-based Auto MAS2026.0525
R-QATModel=Qwen3-1.7B, Quantization Bits=3-bit, Speedup=2.29×2026.0523.7
GADThinking Mode=Non-Thinking2026.0523.34
AMS-ExpectedKV Cache Length (T_keep)=5122026.0523.33
R-KVKV Cache Length (T_keep)=10242026.0523.33
RPCKV Cache Length (T_keep)=10242026.0523.33
ACCBackbone=Qwen3-4B-Base2025.1023.3
rStar-MathBase Model=Qwen3-8B-Base, RM Data=3.64M2026.0523.3
ADASTraining Paradigm=Search-based Auto MAS2026.0523
SCRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-Math-7B2026.0322.8
SFTAccess=text2026.0522.5
SFTBase Model=Qwen2.5-Math-7B, Training Method=SFT2025.0722.3
GRPOBackbone=Qwen3-4B-Base2025.1021.8
Qwen3-4B (student)Thinking Mode=Non-Thinking2026.0520.83
Qwen3-4B (student)Access=–2026.0520.83
Qwen3-8B-BaseRM Data=–2026.0520
AMS-TOVAKV Cache Length (T_keep)=10242026.0520
GPTQModel=R1-Distill-Llama-8B, Quantization Bits=3-bit, Speedup=5.56×2026.0519.64
SATraining Paradigm=Single Agent2026.0519.1
TTRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-Math-7B2026.0319
ParoQModel=Qwen3-1.7B, Quantization Bits=3-bit, Speedup=2.05×2026.0516.88
TTRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-Math-7B2026.0316.8
ScoreFlowTraining Paradigm=RL-based Auto MAS2026.0516.7
MetaAgent-X SFTTraining Paradigm=RL-based Auto MAS2026.0516.7
ChunkKV-ExpectedKV Cache Length (T_keep)=10242026.0516.67
GRPOBase Model=Qwen2.5-Math-7B, Training Method=GRPO2025.0715.3
Re-Schedule_sigmoidBackbone=Qwen2.5-7B, Weight Function=sigmoid2025.1014
RL-PLUSBase Model=Qwen2.5-Math-1.5B, Training Method=RL-PLUS2025.0713.6
AFlowTraining Paradigm=Search-based Auto MAS2026.0513.33
StreamingLLMKV Cache Length (T_keep)=10242026.0513.33