Share your thoughts, 1 month free Claude Pro on usSee more

Logical Reasoning on Countdown

63.2Accuracy

FlowTracer

Updated 1mo ago

Evaluation Results

Method	Links
FlowTracer 2026.06		63.2
Attention 2026.06		60.4
High-entropy 2026.06		57.7
Random 2026.06		55
GRPO 2026.06		52.6
+ EOS Temp. 2026.04		52
Oracle 2026.04		51.3
+ EOS Temp. 2026.04		51.2
Oracle 2026.04		49.6
Auto-ICL 2026.04		49.5
Both 2026.04		48.4
Vanilla 2026.04		48.1
+ Planner 2026.04		46.9
Both 2026.04		46.5
Prob Margin 2026.04		46.1
+ Planner 2026.04		45.7
Prob Margin 2026.04		44.5
Temperature 2026.04		43.8
Auto-ICL 2026.04		42.2
Top1 Prob 2026.04		41.8
Top1 Prob 2026.04		41.4
Temperature 2026.04		41.4
Init. Position 2026.04		39.5
Init. Position 2026.04		36.7
Prefix 2026.04		34.7
Prefix 2026.04		32
Vanilla 2026.04		31
Ancestral 2026.04		25.4
Ancestral 2026.04		23.8
Random 2026.04		21.8
Random 2026.04		17