MERaLiON-2-10B: benchmark results

Provider: A*STAR. Access: Open.

Unified ELO 1552 ± 18, rank #592 of 1607 rated models, from 163 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
SEA-HELM (Indonesian) - Summarization19.28Normalized Score100
SEA-SpeechBench - Emotion Recognition (SEA Prompt)20.34Judge-based accuracy (%; closed nine-class emotion label fro92.9
SEA-SpeechBench - Speech Translation (SEA Prompt)19.52BLEU (0-100): corpus BLEU of the English translation of Sout92.9
SEA-SpeechBench - ASR0.3Word or character error rate (raw ratio, lower is better; WE92.3
SEA-HELM (Indonesian) - NLG55.33Normalized Score86
SEA-HELM (Malay) - Toxicity Detection14.61Normalized Score86
SEA-SpeechBench - Speech Translation (English Prompt)17.75BLEU (0-100): corpus BLEU of the English translation of Sout85.7
SEA-HELM (Filipino) - Summarization21.67Normalized Score80.7
SEA-HELM (Malay) - Safety42.24Normalized Score80.7
SEA-HELM (Vietnamese) - Summarization16.96Normalized Score78.9
SEA-SpeechBench - Temporal Localization (60-120 s)6.4Span-overlap F1 (%; predict the start and end time at which 77.8
SEA-SpeechBench - Speaker Recognition (English Prompt)48.36Macro-F1 (%; whether two clips come from the same speaker, o75

Interactive version: theaggregate.ai/model?slug=meralion-2-10b · How It Works · Data refreshed daily, snapshot 2026-09-29.