Question Answering on ARC-Challenge (Accuracy, TFLOPs)
94.37Accuracydense alignment
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| dense alignmentPair=8B → 14B, Regime=context-aware2026.06 | 94.37 | 21.79 | |
| dense alignmentPair=4B → 14B, Regime=context-aware2026.06 | 94.2 | 21.54 | |
| dense alignmentPair=14B → 8B, Regime=context-aware2026.06 | 93.77 | 15.64 | |
| dense alignmentPair=4B → 8B, Regime=context-aware2026.06 | 93.69 | 12.56 | |
| dense alignmentPair=8B → 4B, Regime=context-aware2026.06 | 93 | 7.95 | |
| Receiver-onlyPair=4B → 14B, Regime=context-aware2026.06 | 92.6 | 31.16 | |
| Receiver-onlyPair=8B → 14B, Regime=context-aware2026.06 | 92.6 | 31.16 | |
| C2CPair=4B → 14B, Regime=context-aware2026.06 | 92.43 | 6.64 | |
| C2CPair=8B → 14B, Regime=context-aware2026.06 | 92.43 | 7.42 | |
| T2TPair=14B → 8B, Regime=context-aware2026.06 | 92.35 | 55.5 | |
| T2TPair=4B → 14B, Regime=context-aware2026.06 | 92 | 56.24 | |
| dense alignmentPair=14B → 4B, Regime=context-aware2026.06 | 91.89 | 10.18 | |
| T2TPair=4B → 8B, Regime=context-aware2026.06 | 91.74 | 37.73 | |
| T2TPair=8B → 14B, Regime=context-aware2026.06 | 91.74 | 67.08 | |
| Receiver-onlyPair=4B → 8B, Regime=context-aware2026.06 | 91 | 19.85 | |
| Receiver-onlyPair=14B → 8B, Regime=context-aware2026.06 | 91 | 19.85 | |
| T2TPair=8B → 4B, Regime=context-aware2026.06 | 90.96 | 33.91 | |
| T2TPair=14B → 4B, Regime=context-aware2026.06 | 89.83 | 43.48 | |
| C2CPair=14B → 8B, Regime=context-aware2026.06 | 89.39 | 7.83 | |
| Receiver-onlyPair=8B → 4B, Regime=context-aware2026.06 | 89.2 | 9.18 | |
| Receiver-onlyPair=14B → 4B, Regime=context-aware2026.06 | 89.2 | 9.18 | |
| C2CPair=8B → 4B, Regime=context-aware2026.06 | 86.78 | 3.43 | |
| C2CPair=4B → 8B, Regime=context-aware2026.06 | 86.09 | 4.5 | |
| C2CPair=14B → 4B, Regime=context-aware2026.06 | 85.83 | 5.01 |