MERaLiON-2-3B: benchmark results
Provider: A*STAR. Access: Open.
Unified ELO 1352 ± 24, rank #1361 of 1632 rated models, from 150 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SEA-SpeechBench - Emotion Recognition (English Prompt) | 23.99 | Judge-based accuracy (%; closed nine-class emotion label fro | 85.7 |
| SEA-SpeechBench - ASR | 0.48 | Word or character error rate (raw ratio, lower is better; WE | 84.6 |
| SEA-SpeechBench - Emotion Recognition (SEA Prompt) | 18.73 | Judge-based accuracy (%; closed nine-class emotion label fro | 71.4 |
| SEA-SpeechBench - Timestamped Content Query (0-30 s) | 4.77 | WER or CER (%; transcribe only the speech inside a queried t | 71.4 |
| SEA-SpeechBench - Temporal Localization (30-60 s) | 10.28 | Span-overlap F1 (%; predict the start and end time at which | 66.7 |
| SEA-SpeechBench - Temporal Localization (0-30 s) | 18.82 | Span-overlap F1 (%; predict the start and end time at which | 64.3 |
| SEA-SpeechBench - Age Recognition (English Prompt) | 33.14 | Macro-F1 (%; speaker age group (teens, adults, seniors) from | 57.1 |
| SEA-SpeechBench - Temporal Localization (60-120 s) | 5.14 | Span-overlap F1 (%; predict the start and end time at which | 55.6 |
| SEA-SpeechBench - Speaker Recognition (English Prompt) | 43.45 | Macro-F1 (%; whether two clips come from the same speaker, o | 50 |
| SEA-SpeechBench - Timestamped Content Query (60-120 s) | 11.27 | WER or CER (%; transcribe only the speech inside a queried t | 44.4 |
| SEA-SpeechBench - Age Recognition (SEA Prompt) | 28.16 | Macro-F1 (%; speaker age group (teens, adults, seniors) from | 35.7 |
| SEA-SpeechBench - Timestamped Content Query (30-60 s) | 8.21 | WER or CER (%; transcribe only the speech inside a queried t | 33.3 |
Interactive version: theaggregate.ai/model?slug=meralion-2-3b · How It Works · Data refreshed daily, snapshot 2026-10-08.