Fact · Transcription
28.87percent word error rate
In the Voxtral paper's FLEURS table, Whisper large-v3 had a word error rate of 28.87 percent in Hindi.
Checked against its source on October 17, 2026
In the source's own words
“Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76”
Voxtral, Mistral AI, arXiv (Audio and Speech Processing), 2025, Table 4 (FLEURS), column hi.
Open the original- What it measures
- FLEURS is a benchmark of read speech. The columns are Arabic (ar), Dutch (nl), English (en), French (fr), German (de), Hindi (hi), Italian (it), Portuguese (pt) and Spanish (es). The table sits in the same paper, so the systems were compared on the same test.
- What it does not show
- One paper's comparison on one test set. Whisper large-v3 is a 2023 model, while the Whisper figures in the 2022 paper are for large-v2, so the two Whisper results are not the same model.