List your company
ResearchFormerly WizWand

Code Generation on CodeContests (Accuracy)

21.2AccuracyAFM-Coder
4.829.072513.32517.5775Feb 26, 2026Mar 11, 2026Mar 25, 2026Apr 8, 2026Apr 22, 2026May 6, 2026May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
AFM-CoderTraining Paradigm=RL-based Auto MAS, Backbone=Qwen3 8B2026.0521.2
SA + GRPOTraining Paradigm=Single Agent2026.0518.6
AGPOclipping=adaptive, ATS=true2026.0517.3
MetaAgent-X RLTraining Paradigm=RL-based Auto MAS, Backbone=Qwen3 8B2026.0517
AFlowTraining Paradigm=Search-based Auto MAS, Backbone=Qwen3 8B2026.0515.8
SATraining Paradigm=Single Agent, Backbone=Qwen3 8B2026.0515.75
AGPOclipping=adaptive, ATS=false2026.0515.2
MaASTraining Paradigm=RL-based Auto MAS, Backbone=Qwen3 8B2026.0515.15
GRPOATS=true2026.0514.9
SATraining Paradigm=Single Agent2026.0514.8
MetaAgent-X RLTraining Paradigm=RL-based Auto MAS2026.0514.2
GRPOclipping=fixed ε2026.0514.1
AFlowTraining Paradigm=Search-based Auto MAS2026.0513.33
ScoreFlowTraining Paradigm=RL-based Auto MAS, Backbone=Qwen3 8B2026.0513.3
Adaptive-KL PPO2026.0513.3
MetaAgent-X SFTTraining Paradigm=RL-based Auto MAS, Backbone=Qwen3 8B2026.0513
DPO2026.0512.4
PPO2026.0512.3
ADASTraining Paradigm=Search-based Auto MAS, Backbone=Qwen3 8B2026.0512.2
ADASTraining Paradigm=Search-based Auto MAS2026.0512.2
SA + GRPOTraining Paradigm=Single Agent, Backbone=Qwen3 8B2026.0512.12
ScoreFlowTraining Paradigm=RL-based Auto MAS2026.0511.92
MaASTraining Paradigm=RL-based Auto MAS2026.059.1
Single AgentBackbone=Qwen3-14B2026.027.27
Dynamic-MAS + ADv2Backbone=Qwen3-14B2026.027.27
Dynamic-MASBackbone=Qwen3-14B2026.026.67
Dynamic-MAS + Self-RefineBackbone=Qwen3-14B2026.026.06
Dynamic-MAS + Multi-TAGBackbone=Qwen3-14B2026.026.06
MetaAgent-X SFTTraining Paradigm=RL-based Auto MAS2026.056
Dynamic-MAS + PRMBackbone=Qwen3-14B2026.025.45