model detail

Where this model stands against the best model in each language at Q8_0, its speed on an AMD Ryzen 4750U laptop, and what the card says. Add a second model to compare them head to head.

model whisper-large
AMD 4750U GPU2.0×
AMD 4750U CPU1.0×
Q8_0 size1.55 GB
frontier · AMD 4750U GPU1 of 82

verdict: Never first, but tied with the leader in 4 of 82 languages (overlapping 95% CIs). Weakest relative to the field in Maltese (4.9× the leader's error), Hindi (3.9×), and Japanese (CER) (3.0×). Fast enough for realtime on the AMD 4750U GPU (2.0×), not on the AMD 4750U CPU (1.0×).

license: Apache-2.0 · card languages: 102 · translatelang detecttimestamps (segment)

standing by language · 82 benchmarked

Each bar compares this model with the best other model in that language, on a log scale: left of center it is ahead, right of center behind, and the other model is named on the row. The two error rates use the legend colors: this model first, the other model second. Faded bars are ties, where the 95% confidence intervals overlap. Sorted strongest standing first. Hover a row for full details.

whisper-large error best other model error tie: the 95% CIs overlap

quantization · 23 usable languages at Q5_K_M

Q5_K_M is within noise on every one of 23 usable languages: 70% of the Q8_0 file for no measurable accuracy cost.

Q5_K_M file70% of Q8_01.08 GB against 1.55 GB
median error change+0.8%over 23 usable languages
measurably worse0 of 23above the Q8_0 95% confidence interval
worst usable languageGreek18.79 → 19.67 (+0.88 pts)
measurably worse than Q8_0: above its 95% confidence interval within noise typical noise of one measurement
per-language numbers (23 usable, 22 more at 30% or worse)
Greek 18.7917.47–20.14 19.67+0.88 +4.7%
Italian 3.673.26–4.11 3.83+0.16 +4.4%
Spanish 3.553.18–3.98 3.63+0.08 +2.3%
Hungarian 21.9721.06–22.93 22.34+0.37 +1.7%
Bulgarian 19.8918.96–20.85 20.20+0.31 +1.6%
Czech 18.7717.92–19.75 19.04+0.27 +1.4%
Korean (CER) 5.254.26–6.34 5.32+0.07 +1.3%
Hindi 29.1027.16–31.25 29.46+0.36 +1.2%
Filipino 15.4414.78–16.15 15.62+0.18 +1.2%
Japanese (CER) 6.956.39–7.50 7.01+0.06 +0.9%
Bosnian 22.2021.32–23.09 22.38+0.18 +0.8%
Croatian 18.3217.52–19.12 18.46+0.14 +0.8%
Catalan 7.016.46–7.61 7.06+0.05 +0.7%
English 4.463.90–5.11 4.49+0.03 +0.7%
Finnish 12.7312.01–13.53 12.81+0.08 +0.6%
Macedonian 22.8121.91–23.68 22.95+0.14 +0.6%
Estonian 27.3126.31–28.35 27.45+0.14 +0.5%
German 5.044.63–5.47 5.06+0.02 +0.4%
French 6.956.39–7.59 6.96+0.01 +0.1%
Arabic 19.4718.01–21.12 19.42−0.05 −0.3%
Danish 17.8216.95–18.65 17.77−0.05 −0.3%
Galician 20.1019.39–20.93 20.01−0.09 −0.4%
Indonesian 8.938.27–9.68 8.78−0.15 −1.7%

speed · all 73 models on the AMD 4750U GPU

× realtime, log scale. Left of the realtime marker, transcribing takes longer than the recording. Hover a dot for the model.

compare with

whisper-basesame familywhisper-large-v2same familywhisper-large-v3same familywhisper-large-v3-turbosame familyparakeet-ultraahead of it in 23 of 25 shared languageswhisper-mediumclosest coverage, 82 languages in common