System-level Quality Estimation Correlation on WMT Czech → German 2025 (test)
91Soft Pairwise AccuracyGemini1.5_flash-ESA
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini1.5_flash-ESAMetric Category=LLM-based, Scoring Protocol=Explicit Score Assignment (ESA), Model Backbone=Gemini 1.5 Flash2026.05 | 91 | |
| Gemini1.5-ProMetric Category=LLM-based, Model Backbone=Gemini 1.5 Pro2026.05 | 90 | |
| GEMBA-v2Metric Category=LLM-based2026.05 | 84 | |
| Qwen-ESAMetric Category=LLM-based, Scoring Protocol=Explicit Score Assignment (ESA), Model Backbone=Qwen2026.05 | 80 | |
| Gemma-ESAMetric Category=LLM-based, Scoring Protocol=Explicit Score Assignment (ESA), Model Backbone=Gemma2026.05 | 80 | |
| COMET22Metric Category=Baseline2026.05 | 77 | |
| EuroLLM-ESAMetric Category=LLM-based, Scoring Protocol=Explicit Score Assignment (ESA), Model Backbone=EuroLLM2026.05 | 76 | |
| MetricX-24-QEMetric Category=Baseline2026.05 | 72 | |
| COMETKiwi22Metric Category=Baseline2026.05 | 61 |