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-v2
AMD 4750U GPU2.0×
AMD 4750U CPU1.0×
Q8_0 size1.55 GB
frontier · AMD 4750U GPU8 of 82

verdict: Best in 2 of 82 languages, and tied with the leader in 13 more (overlapping 95% CIs). Weakest relative to the field in Maltese (4.3× the leader's error), Hindi (3.1×), and Romanian (2.5×). 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-v2 error best other model error tie: the 95% CIs overlap

quantization · 26 usable languages at Q5_K_M

Q5_K_M is within noise on every one of 26 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+0.6%over 26 usable languages
measurably worse0 of 26above the Q8_0 95% confidence interval
worst usable languageFrench5.81 → 6.01 (+0.20 pts)
measurably worse than Q8_0: above its 95% confidence interval within noise typical noise of one measurement
per-language numbers (26 usable, 19 more at 30% or worse)
French 5.815.29–6.33 6.01+0.20 +3.4%
Greek 13.5112.53–14.42 13.81+0.30 +2.2%
Hindi 23.2721.44–25.21 23.71+0.44 +1.9%
Indonesian 7.436.82–8.12 7.55+0.12 +1.6%
Croatian 14.1813.43–14.97 14.39+0.21 +1.5%
Danish 14.9214.19–15.63 15.14+0.22 +1.5%
Filipino 13.1712.48–13.88 13.36+0.19 +1.4%
English 4.213.73–4.82 4.27+0.06 +1.4%
Hungarian 17.8417.00–18.71 18.07+0.23 +1.3%
Czech 14.4213.62–15.22 14.59+0.17 +1.2%
Galician 16.5715.91–17.25 16.73+0.16 +1.0%
Macedonian 18.7617.94–19.58 18.92+0.16 +0.9%
Arabic 17.0615.68–18.51 17.17+0.11 +0.6%
Hebrew 27.7826.13–29.34 27.91+0.13 +0.5%
Estonian 23.2522.23–24.25 23.33+0.08 +0.3%
Bulgarian 15.8115.00–16.69 15.86+0.05 +0.3%
Finnish 9.588.93–10.26 9.61+0.03 +0.3%
German 4.534.13–4.93 4.54+0.01 +0.2%
Japanese (CER) 5.565.09–6.06 5.57+0.01 +0.2%
Azerbaijani 24.1323.27–24.98 24.13+0.00 +0.0%
Spanish 3.302.93–3.70 3.30+0.00 +0.0%
Bosnian 17.0216.25–17.87 17.00−0.02 −0.1%
Latvian 24.4923.58–25.48 24.42−0.07 −0.3%
Korean (CER) 4.993.99–6.00 4.97−0.02 −0.4%
Catalan 5.565.03–6.12 5.52−0.04 −0.7%
Italian 3.593.07–4.21 3.52−0.07 −1.9%

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-largesame familywhisper-large-v3same familywhisper-large-v3-turbosame familycanary-1b-v2ahead of it in 18 of 25 shared languageswhisper-mediumclosest coverage, 82 languages in common