List your company

Mathematical Reasoning on GSM8K matched-N groups (test)

96.33AccuracyOracle topology
93.906894.535995.16595.7941May 28, 2026

Evaluation results

MethodLinks
Oracle topologyTopology Category=Confidence-routed2026.0596.33
hub-spokeTopology Category=Fixed topology2026.0595.33
Per-bench best fixedTopology Category=Fixed topology2026.0595.33
Majority over topologiesTopology Category=Selection w/o learned confidence2026.0595.33
Highest plurality shareTopology Category=Selection w/o learned confidence2026.0595.33
CAGE-SELECTTopology Category=Confidence-routed2026.0595.33
treeTopology Category=Fixed topology2026.0595
iidTopology Category=Fixed topology2026.0594.67
debateTopology Category=Fixed topology2026.0594.67
Highest mean log-probTopology Category=Selection w/o learned confidence2026.0594.67
chainTopology Category=Fixed topology2026.0594