List your company

Active Reasoning on ARBench SP

43.1Accuracyo4-mini
34.88437.01739.1541.283May 1, 2026

Evaluation results

MethodLinks
o4-miniSize=–2026.0543.1
TeamTR (Homogeneous)Size=3×8B2026.0543.1
TeamTR (Heterogeneous)Size=1.7B+8B+14B2026.0542.3
GPT-4o-miniSize=–2026.0540.8
Debate + JudgeSize=3×8B2026.0539.4
Qwen3 (thinking)Size=32B2026.0538.9
Role-playSize=3×8B2026.0538.7
Qwen3-A3BSize=30B2026.0538.4
DebateSize=3×8B2026.0538
QwQSize=32B2026.0536.1
DAPOSize=8B2026.0535.9
DeepSeek-R1 DistillSize=70B2026.0535.4
GRPOSize=8B2026.0535.2