List your company

Active Reasoning on ARBench-GN

45.3AccuracyTeamTR (Homogeneous)
38.64440.37242.143.828May 1, 2026

Evaluation results

MethodLinks
TeamTR (Homogeneous)Size=3×8B2026.0545.3
o4-miniSize=–2026.0545.1
TeamTR (Heterogeneous)Size=1.7B+8B+14B2026.0544.5
GPT-4o-miniSize=–2026.0543.6
Debate + JudgeSize=3×8B2026.0543.5
Role-playSize=3×8B2026.0543.1
Qwen3-A3BSize=30B2026.0542.3
DebateSize=3×8B2026.0542.3
Qwen3 (thinking)Size=32B2026.0541.1
DeepSeek-R1 DistillSize=70B2026.0540.7
QwQSize=32B2026.0539.5
DAPOSize=8B2026.0539.3
GRPOSize=8B2026.0538.9