System-level Speech Translation Quality Estimation on IWSLT Metrics Shared Task 2026 (dev)
99.7Soft Pairwise Accuracy (en-de)HydraQE (DA head)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HydraQE (DA head)2026.06 | 99.7 | 69.2 | 84.5 | |
| MetricX-24-XXL QE2026.06 | 99.6 | 91.3 | 95.5 | |
| HydraQE (all heads avg.)2026.06 | 99.5 | 56 | 77.7 | |
| HydraQE (primary)2026.06 | 99.4 | 79.3 | 89.4 | |
| HydraQE (MetricX head)2026.06 | 98.8 | 90.1 | 94.5 | |
| HydraQE (xCOMET head)2026.06 | 96.9 | 34.6 | 65.7 | |
| CometKiwivariant=XXL2026.06 | 91.6 | 34.9 | 63.3 | |
| xCOMET-XXL QE2026.06 | 87.7 | 45.8 | 66.8 | |
| COMET2026.06 | 86.2 | 92.6 | 89.4 | |
| BLASER 2.0 QE2026.06 | 86 | 67.7 | 76.9 | |
| SpeechQE2026.06 | 78.6 | 93.4 | 86 | |
| COMET (partial)2026.06 | 44.4 | 68.7 | 56.6 |