language benchmark
Translation Set1→en COMET22
COMET-22 is a neural machine translation evaluation metric that uses an ensemble of two models: a COMET estimator trained with Direct Assessments and a multitask model that predicts sentence-level scores and word-level OK/BAD tags. It provides improved correlations with human judgments and increased robustness to critical errors compared to previous metrics.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
Translation Set1→en COMET22 leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AM Nova ProAmazon | 89.0% | 100.0% | 3 | C | |
| 2 | AMNova LiteAmazon | 88.8% | 50.0% | 3 | C | |
| 3 | AMNova MicroAmazon | 88.7% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
Translation Set1→en COMET22 highlights
The top published results on this benchmark's own scale.
What is Translation Set1→en COMET22?
Definition and scoring fields from the benchmark registry.
COMET-22 is a neural machine translation evaluation metric that uses an ensemble of two models: a COMET estimator trained with Direct Assessments and a multitask model that predicts sentence-level scores and word-level OK/BAD tags. It provides improved correlations with human judgments and increased robustness to critical errors compared to previous metrics.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Translation Set1→en COMET22
- Modality
- text
- Primary category
- language
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- translation-set1→en-comet22|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Translation Set1→en COMET22.
Which model scores highest on Translation Set1→en COMET22?
Nova Pro is currently ranked first with 89.0%.
What does Translation Set1→en COMET22 measure?
COMET-22 is a neural machine translation evaluation metric that uses an ensemble of two models: a COMET estimator trained with Direct Assessments and a multitask model that predicts sentence-level scores and word-level OK/BAD tags. It provides improved correlations with human judgments and increased robustness to critical errors compared to previous metrics.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.