List your company
ResearchFormerly WizWand

Accuracy on GPQA

71.43AccuracyGRPO
9.227625.376341.52557.6737May 8, 2025Jul 13, 2025Sep 17, 2025Nov 22, 2025Jan 27, 2026Apr 3, 2026Jun 9, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GRPOInference Setting=Scaffold-free2026.0671.43
ParaBridgeInference Setting=Scaffold-free2026.0671.43
BaselineInference Setting=Scaffold-free2026.0671.34
RFTInference Setting=Scaffold-free2026.0668.45
Claude 3.5 SonnetEvaluation Protocol=0-shot CoT2026.0359.4
DyLANBase Model=gpt-4o, Number of Agents=82026.0458.89
HuggingGPTImplementation Framework=original2026.0356.67
GoAMaxBase Model=gpt-4o, Number of Agents=62026.0456.57
Layer-wise CapreseModel Size=14B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0555.05
GoAMaxBase Model=gpt-4o, Number of Agents=32026.0455.05
RefineBase Model=gpt-4o, Number of Agents=62026.0454.98
SCBase Model=gpt-4o, Number of Agents=62026.0454.27
GPT-4oEvaluation Protocol=0-shot CoT2026.0353.6
DebateBase Model=gpt-4o, Number of Agents=62026.0453.03
ReconcileBase Model=gpt-4o, Number of Agents=62026.0453.03
DS-Qwen 14BModel Size=14B, 0-shot=true2025.0552.53
Llama 3 405BEvaluation Protocol=0-shot CoT2026.0351.1
E2E CapreseModel Size=14B, Sparse Algorithm=CATS, 0-shot=true2025.0551.01
Layer-wise CapreseModel Size=14B, Sparse Algorithm=CATS, 0-shot=true2025.0550.51
MoABase Model=gpt-4o, Number of Agents=62026.0450.51
AlphaRL-enhanced DAPOAlgorithm=DAPO, Stage=40%2025.1049.25
DAPOAlgorithm=DAPO, Stage=Fully Trained2025.1048.23
Qwen3-4B + TTSVModel=Qwen3-4B, Adaptation=TTSV2025.1247.98
gpt-4oBase Model=gpt-4o, Number of Agents=12026.0447.47
HuggingGPTImplementation Framework=MASFactory2026.0347.32
GRPOAlgorithm=GRPO, Stage=Fully Trained2025.1047.1
Llama 3 70BEvaluation Protocol=0-shot CoT2026.0346.7
Qwen3-4B-Instruct-2507Model=Qwen3-4B-Instruct-2507, Training Method=Instruct2026.0346.6
RLOOAlgorithm=RLOO, Stage=Fully Trained2025.1045.82
HeRLModel=Qwen3-4B-Instruct-2507, Training Method=HeRL2026.0345.3
CATSModel Size=14B, 0-shot=true2025.0544.95
AlphaRL-enhanced RLOOAlgorithm=RLOO, Stage=10%2025.1044.95
AlphaRL-enhanced RLOOAlgorithm=RLOO, Stage=40%2025.1044.4
Relational AggregationCalls=3, Tokens(k)=0.663, Time(s)=0.202026.0543.64
Relational AggregationCategory=Ours2026.0543.64
E2E CapreseModel Size=14B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0543.43
AlphaRL-enhanced GRPOAlgorithm=GRPO, Stage=40%2025.1043.13
Anchor (Refined Belief)Category=Ours2026.0542.14
RLOOAlgorithm=RLOO, Stage=40%2025.1042.05
Qwen3-4BModel=Qwen3-4B2025.1241.92
GRIFFINModel Size=14B, 0-shot=true2025.0541.92
DAPOAlgorithm=DAPO, Stage=40%2025.1041.67
AlphaRL-enhanced DAPOAlgorithm=DAPO, Stage=10%2025.1041.54
GRPOAlgorithm=GRPO, Stage=40%2025.1041.16
Anchor (Initial Belief)Category=Ours2026.0540.61
GoAMeanCategory=Multi-Agent2026.0540.54
GoA-MaxCalls=11, Tokens(k)=17.32, Time(s)=88.152026.0539.98
GoAMaxCategory=Multi-Agent2026.0539.98
Vibe Graphing-Task SpecificImplementation Framework=Vibe Graphing2026.0339.51
Majority (Refined Belief)Category=Ours2026.0539.41
CharacterFlywheel V7Evaluation Protocol=0-shot CoT2026.0339.3
RefineCategory=Multi-Agent2026.0538.92
RLOOAlgorithm=RLOO, Stage=10%2025.1038.65
AgentVerseImplementation Framework=original2026.0338.39
DS-Qwen 7BModel Size=7B, 0-shot=true2025.0538.38
Layer-wise CapreseModel Size=7B, Sparse Algorithm=CATS, 0-shot=true2025.0538.38
AgentVerseImplementation Framework=MASFactory2026.0337.5
Majority (Initial Belief)Category=Ours2026.0537.26
GRPOAlgorithm=GRPO, Stage=10%2025.1036.74
AlphaRL-enhanced GRPOAlgorithm=GRPO, Stage=10%2025.1036.74
DAPOAlgorithm=DAPO, Stage=10%2025.1036.66
SCCategory=Multi-Agent2026.0536.36
E2E CapreseModel Size=7B, Sparse Algorithm=CATS, 0-shot=true2025.0535.86
ReConcileCategory=Multi-Agent2026.0534.34
Qwen2.5-7B-InstructModel=Qwen2.5-7B-Instruct, Training Method=Instruct2026.0334.1
HeRLModel=Qwen2.5-7B-Instruct, Training Method=HeRL2026.0333.9
CodeCategory=Single-Agent2026.0533.84
Self-MoACategory=Multi-Agent2026.0533.84
CATSModel Size=7B, 0-shot=true2025.0532.83
MoACalls=19, Tokens(k)=56.87, Time(s)=245.652026.0532.83
GeneralCategory=Single-Agent2026.0532.83
MoACategory=Multi-Agent2026.0532.83
Llama 3 8BEvaluation Protocol=0-shot CoT2026.0332.8
CAMELImplementation Framework=original2026.0332.59
MathCategory=Single-Agent2026.0530.81
LegalCategory=Single-Agent2026.0530.3
DebateCategory=Multi-Agent2026.0529.29
HeRLModel=Llama-3.2-3B-Instruct, Training Method=HeRL2026.0329.2
FinanceCategory=Single-Agent2026.0528.28
Layer-wise CapreseModel Size=7B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0527.78
Llama-3.2-3B-InstructModel=Llama-3.2-3B-Instruct, Training Method=Instruct2026.0326.5
BiomedicalCategory=Single-Agent2026.0525.25
CAMELImplementation Framework=MASFactory2026.0324.78
E2E CapreseModel Size=7B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0523.74
E2E CapreseModel Size=1.5B, Sparse Algorithm=CATS, 0-shot=true2025.0522.22
GRIFFINModel Size=7B, 0-shot=true2025.0521.72
DS-Qwen 1.5BModel Size=1.5B, 0-shot=true2025.0518.69
E2E CapreseModel Size=1.5B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0516.67
Layer-wise CapreseModel Size=1.5B, Sparse Algorithm=CATS, 0-shot=true2025.0516.67
Layer-wise CapreseModel Size=1.5B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0513.13
GRIFFINModel Size=1.5B, 0-shot=true2025.0511.62
CATSModel Size=1.5B, 0-shot=true2025.0511.62