Question Answering on MMLU-Pro matched-N groups (test)
65.31AccuracyOracle topology
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle topologyTopology Category=Confidence-routed2026.05 | 65.31 | |
| CAGE-SELECTTopology Category=Confidence-routed2026.05 | 50.34 | |
| debateTopology Category=Fixed topology2026.05 | 46.94 | |
| Per-bench best fixedTopology Category=Fixed topology2026.05 | 46.94 | |
| chainTopology Category=Fixed topology2026.05 | 46.26 | |
| Highest plurality shareTopology Category=Selection w/o learned confidence2026.05 | 44.56 | |
| hub-spokeTopology Category=Fixed topology2026.05 | 42.86 | |
| iidTopology Category=Fixed topology2026.05 | 41.5 | |
| treeTopology Category=Fixed topology2026.05 | 41.5 | |
| Majority over topologiesTopology Category=Selection w/o learned confidence2026.05 | 40.82 | |
| Highest mean log-probTopology Category=Selection w/o learned confidence2026.05 | 40.82 |