Whisper Large-v3: benchmark results

Provider: Other. Released 2023-11-06. Access: Open.

Unified ELO 1422 ± 31, rank #1729 of 2656 rated models, from 30 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AudioBench - GigaSpeech WER9.46WER (%) (lower is better)100
AudioBench - IMDA Part 3 ASR WER27.03WER (%) (lower is better)100
AudioBench - LibriSpeech Other WER3.66WER (%) (lower is better)100
AudioBench - TED-LIUM 3 Long Form WER3.21WER (%) (lower is better)100
ProLLM - Transcription77.9Score (%)100
AudioBench - IMDA Part 5 ASR WER21.44WER (%) (lower is better)88.9
AudioBench - TED-LIUM 3 WER3.76WER (%) (lower is better)88.9
AudioBench - YouTube ASR Batch 2 WER17.21WER (%) (lower is better)87.5
AudioBench - SEAME Singapore English WER53.77WER (%) (lower is better)80
AudioBench - Earnings21 WER11.86WER (%) (lower is better)77.8
AudioBench - Earnings22 WER15.89WER (%) (lower is better)77.8
AudioBench - IMDA Part 1 ASR WER6.84WER (%) (lower is better)77.8

Interactive version: theaggregate.ai/model?slug=whisper-large-v3 · How It Works · Data refreshed daily, snapshot 2026-09-19.