Think Again or Think Longer? Selective Verification for Budget-Aware Reasoning
Open paper
Benchmark
Model
Rank
Results
arithmetic-reasoning-on-gsm8k
Selective active verify (Qwen3-1.7B gate)
#29
Accuracy: 94.47
mathematical-reasoning-on-math-500
Selective active verify (Qwen3-1.7B gate)
#104
Accuracy: 76.3
Rank counts only results with a code link.