handy models

speech-to-text benchmarks for transcribe.cpp, the engine behind Handy

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 Voxtral-Mini-3B-2507
AMD 4750U GPU1.1×
AMD 4750U CPU0.6×
Q8_0 size4.66 GB
frontier · AMD 4750U GPU2 of 8

verdict: Never first, but tied with the leader in 6 of 8 languages (overlapping 95% CIs). Weakest relative to the field in German (1.4× the leader's error) and Spanish (1.3×). Fast enough for realtime on the AMD 4750U GPU (1.1×), not on the AMD 4750U CPU (0.6×).

license: Apache-2.0 · card languages: 8 · translatelang detect

standing by language · 8 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.

Voxtral-Mini-3B-2507 error best other model error tie: the 95% CIs overlap

quantization · 8 usable languages at Q5_K_M

Q5_K_M is within noise on every one of 8 usable languages: 69% of the Q8_0 file for no measurable accuracy cost.

Q5_K_M file69% of Q8_03.23 GB against 4.66 GB
median error change+1.9%over 8 usable languages
measurably worse0 of 8above the Q8_0 95% confidence interval
worst usable languageFrench4.51 → 5.01 (+0.50 pts)
measurably worse than Q8_0: above its 95% confidence interval within noise typical noise of one measurement
per-language numbers (8)
French 4.514.04–5.02 5.01+0.50 +11.1%
Italian 2.562.24–2.90 2.76+0.20 +7.8%
Dutch 6.575.84–7.32 6.95+0.38 +5.8%
Portuguese 3.843.44–4.31 3.95+0.11 +2.9%
Spanish 3.523.11–3.95 3.55+0.03 +0.9%
German 4.714.26–5.21 4.74+0.03 +0.6%
Hindi 8.938.13–9.91 8.95+0.02 +0.2%
English 3.893.47–4.35 3.87−0.02 −0.5%

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

Voxtral-Mini-4B-Realtime-2602same familyVoxtral-Small-24B-2507same familyQwen3-ASR-1.7Bahead of it in 4 of 8 shared languagesgranite-speech-4.1-2b-narclosest coverage, 5 languages in common