recommended models with English

English first, then 50 English-language pairs sorted by the overall pick’s accuracy/speed score, best first. Each cell shows the pick’s error on its worst language; hover for the combined error, wait and score. 9 rows are experimental.

overall accuracy/speed balance within the wait target on the 4750U GPUfast most accurate within the wait target on the 4750U CPUaccurate most accurate within the wait target on the 4750U GPU
languages overall≤5s GPU fast≤4s CPU accurate≤10s GPU
English en parakeet-unified-en-0.6b 3.99% 1.2s GPU wait parakeet-unified-en-0.6b 3.99% 2.1s CPU wait Qwen3-ASR-1.7B 3.23% 7.3s GPU wait
English + Italian en + it parakeet-ultra 4.26% 1.3s GPU wait parakeet-ultra 4.26% 2.3s CPU wait Qwen3-ASR-1.7B 3.23% 7.3s GPU wait
English + Spanish en + es parakeet-ultra 4.26% 1.3s GPU wait parakeet-ultra 4.26% 2.3s CPU wait Qwen3-ASR-1.7B 3.31% 7.3s GPU wait
English + Portuguese en + pt parakeet-ultra 4.26% 1.3s GPU wait parakeet-ultra 4.26% 2.3s CPU wait Qwen3-ASR-1.7B 4.37% 7.3s GPU wait
English + German en + de parakeet-ultra 4.58% 1.3s GPU wait parakeet-ultra 4.58% 2.3s CPU wait Qwen3-ASR-1.7B 4.25% 7.3s GPU wait
English + French en + fr parakeet-ultra 4.66% 1.3s GPU wait parakeet-ultra 4.66% 2.3s CPU wait Qwen3-ASR-1.7B 4.52% 7.3s GPU wait
English + Russian en + ru parakeet-ultra 5.63% 1.3s GPU wait parakeet-ultra 5.63% 2.3s CPU wait Qwen3-ASR-1.7B 6.25% 7.3s GPU wait
English + Ukrainian en + uk parakeet-ultra 5.71% 1.3s GPU wait parakeet-ultra 5.71% 2.3s CPU wait parakeet-ultra 5.71% 1.3s GPU wait
English + Korean en + ko Qwen3-ASR-0.6B 5.82% 3.1s GPU wait SenseVoiceSmall 8.27% 1.5s CPU wait Qwen3-ASR-1.7B 4.60% 7.3s GPU wait
English + Polish en + pl parakeet-ultra 6.16% 1.3s GPU wait parakeet-ultra 6.16% 2.3s CPU wait whisper-large-v3-turbo 5.81% 9.7s GPU wait
English + Dutch en + nl parakeet-ultra 6.70% 1.3s GPU wait parakeet-ultra 6.70% 2.3s CPU wait Qwen3-ASR-1.7B 7.43% 7.3s GPU wait
English + Mandarin Chinese en + zh Qwen3-ASR-0.6B 7.57% 3.1s GPU wait SenseVoiceSmall 10.12% 1.5s CPU wait Qwen3-ASR-1.7B 7.14% 7.3s GPU wait
English + Slovak en + sk parakeet-ultra 7.83% 1.3s GPU wait parakeet-ultra 7.83% 2.3s CPU wait parakeet-ultra 7.83% 1.3s GPU wait
English + Cantonese en + yue Qwen3-ASR-0.6B 7.91% 3.1s GPU wait Qwen3-ASR-0.6B 7.91% 5.7s CPU wait over 4s target Qwen3-ASR-1.7B 6.13% 7.3s GPU wait
English + Indonesian en + id Qwen3-ASR-0.6B 8.49% 3.1s GPU wait whisper-small 18.02% 5.1s CPU wait over 4s target Qwen3-ASR-1.7B 5.37% 7.3s GPU wait
English + Japanese en + ja Qwen3-ASR-0.6B 8.61% 3.1s GPU wait SenseVoiceSmall 7.63% 1.5s CPU wait Qwen3-ASR-1.7B 5.29% 7.3s GPU wait
English + Thai en + th Qwen3-ASR-0.6B 8.81% 3.1s GPU wait Qwen3-ASR-0.6B 8.81% 5.7s CPU wait over 4s target Qwen3-ASR-1.7B 6.89% 7.3s GPU wait
English + Vietnamese en + vi Qwen3-ASR-0.6B 9.32% 3.1s GPU wait nemotron-3.5-asr-streaming-0.6b 13.96% 2.8s CPU wait Qwen3-ASR-1.7B 6.15% 7.3s GPU wait
English + Romanian en + ro parakeet-ultra 10.12% 1.3s GPU wait parakeet-ultra 10.12% 2.3s CPU wait parakeet-ultra 10.12% 1.3s GPU wait
English + Czech en + cs orukeet 10.22% 1.2s GPU wait orukeet 10.22% 2.2s CPU wait orukeet 10.22% 1.2s GPU wait
English + Bulgarian en + bg parakeet-ultra 11.04% 1.3s GPU wait parakeet-ultra 11.04% 2.3s CPU wait parakeet-ultra 11.04% 1.3s GPU wait
English + Finnish en + fi parakeet-ultra 11.14% 1.3s GPU wait parakeet-ultra 11.14% 2.3s CPU wait whisper-large-v3-turbo 8.29% 9.7s GPU wait
English + Croatian en + hr parakeet-ultra 11.30% 1.3s GPU wait parakeet-ultra 11.30% 2.3s CPU wait parakeet-ultra 11.30% 1.3s GPU wait
English + Hindi en + hi Qwen3-ASR-0.6B 12.68% 3.1s GPU wait nemotron-3.5-asr-streaming-0.6b 8.61% 2.8s CPU wait Qwen3-ASR-1.7B 7.84% 7.3s GPU wait
English + Swedish en + sv parakeet-ultra 12.83% 1.3s GPU wait parakeet-ultra 12.83% 2.3s CPU wait whisper-large-v3-turbo 8.72% 9.7s GPU wait
English + Hungarian en + hu parakeet-ultra 13.09% 1.3s GPU wait parakeet-ultra 13.09% 2.3s CPU wait parakeet-ultra 13.09% 1.3s GPU wait
English + Estonian en + et parakeet-ultra 14.44% 1.3s GPU wait parakeet-ultra 14.44% 2.3s CPU wait parakeet-ultra 14.44% 1.3s GPU wait
English + Danish en + da parakeet-ultra 15.98% 1.3s GPU wait parakeet-ultra 15.98% 2.3s CPU wait whisper-large-v3-turbo 13.60% 9.7s GPU wait
English + Turkish en + tr Qwen3-ASR-0.6B 16.74% 3.1s GPU wait nemotron-3.5-asr-streaming-0.6b 15.40% 2.8s CPU wait whisper-large-v3-turbo 6.97% 9.7s GPU wait
English + Maltese en + mt parakeet-ultra 17.01% 1.3s GPU wait parakeet-ultra 17.01% 2.3s CPU wait parakeet-ultra 17.01% 1.3s GPU wait
English + Lithuanian en + lt parakeet-ultra 17.06% 1.3s GPU wait parakeet-ultra 17.06% 2.3s CPU wait parakeet-ultra 17.06% 1.3s GPU wait
English + Malay en + ms Qwen3-ASR-0.6B 17.18% 3.1s GPU wait Qwen3-ASR-0.6B 17.18% 5.7s CPU wait over 4s target Qwen3-ASR-1.7B 10.42% 7.3s GPU wait
English + Latvian en + lv parakeet-ultra 18.04% 1.3s GPU wait parakeet-ultra 18.04% 2.3s CPU wait parakeet-ultra 18.04% 1.3s GPU wait
English + Slovenian en + sl parakeet-ultra 18.98% 1.3s GPU wait parakeet-ultra 18.98% 2.3s CPU wait parakeet-ultra 18.98% 1.3s GPU wait
English + Catalan en + ca whisper-small 14.19% 2.6s GPU wait whisper-small 14.19% 5.1s CPU wait over 4s target whisper-large-v3-turbo 5.42% 9.7s GPU wait
English + Macedonian en + mk Qwen3-ASR-1.7B 18.22% 7.3s GPU wait over 5s target Qwen3-ASR-1.7B 18.22% 12.1s CPU wait over 4s target Qwen3-ASR-1.7B 18.22% 7.3s GPU wait
English + Filipino en + fil whisper-large-v3-turbo 12.08% 9.7s GPU wait over 5s target whisper-medium 18.36% 15.6s CPU wait over 4s target whisper-large-v3-turbo 12.08% 9.7s GPU wait
English + Galician en + gl whisper-large-v3-turbo 12.76% 9.7s GPU wait over 5s target whisper-large-v3-turbo 12.76% 21.0s CPU wait over 4s target whisper-large-v3-turbo 12.76% 9.7s GPU wait
English + Greek en + el whisper-large-v3-turbo 13.26% 9.7s GPU wait over 5s target whisper-large-v3-turbo 13.26% 21.0s CPU wait over 4s target whisper-large-v3-turbo 13.26% 9.7s GPU wait
English + Arabic en + ar nemotron-3.5-asr-streaming-0.6b 15.93% 1.8s GPU wait nemotron-3.5-asr-streaming-0.6b 15.93% 2.8s CPU wait Qwen3-ASR-1.7B 14.91% 7.3s GPU wait
English + Bosnian en + bs whisper-large-v3-turbo 14.77% 9.7s GPU wait over 5s target whisper-large-v3-turbo 14.77% 21.0s CPU wait over 4s target whisper-large-v3-turbo 14.77% 9.7s GPU wait
English + Persian en + fa experimental Qwen3-ASR-1.7B 28.29% 7.3s GPU wait over 5s target Qwen3-ASR-1.7B 28.29% 12.1s CPU wait over 4s target Qwen3-ASR-1.7B 28.29% 7.3s GPU wait
English + Norwegian Bokmal en + nb nemotron-3.5-asr-streaming-0.6b 19.24% 1.8s GPU wait nemotron-3.5-asr-streaming-0.6b 19.24% 2.8s CPU wait whisper-large-v3-turbo 9.10% 9.7s GPU wait
English + Kazakh en + kk experimental whisper-large-v3-turbo 21.27% 9.7s GPU wait over 5s target whisper-large-v3-turbo 21.27% 21.0s CPU wait over 4s target whisper-large-v3-turbo 21.27% 9.7s GPU wait
English + Icelandic en + is experimental whisper-large-v3-turbo 21.39% 9.7s GPU wait over 5s target whisper-large-v3-turbo 21.39% 21.0s CPU wait over 4s target whisper-large-v3-turbo 21.39% 9.7s GPU wait
English + Tamil en + ta experimental whisper-large-v3 20.04% 15.0s GPU wait over 5s target whisper-large-v3 20.04% 30.9s CPU wait over 4s target whisper-large-v3 20.04% 15.0s GPU wait over 10s target
English + Azerbaijani en + az experimental whisper-large-v3 21.10% 15.0s GPU wait over 5s target whisper-large-v3 21.10% 30.9s CPU wait over 4s target whisper-large-v3 21.10% 15.0s GPU wait over 10s target
English + Urdu en + ur experimental whisper-large-v3 21.75% 15.0s GPU wait over 5s target whisper-large-v3 21.75% 30.9s CPU wait over 4s target whisper-large-v3 21.75% 15.0s GPU wait over 10s target
English + Hebrew en + he experimental whisper-large-v3 26.73% 15.0s GPU wait over 5s target whisper-large-v3 26.73% 30.9s CPU wait over 4s target whisper-large-v3 26.73% 15.0s GPU wait over 10s target
English + Serbian en + sr experimental whisper-large-v3 28.49% 15.0s GPU wait over 5s target whisper-large-v3 28.49% 30.9s CPU wait over 4s target whisper-large-v3 28.49% 15.0s GPU wait over 10s target
English + Welsh en + cy experimental whisper-large-v2 30.55% 14.9s GPU wait over 5s target whisper-large-v2 30.55% 31.3s CPU wait over 4s target whisper-large-v2 30.55% 14.9s GPU wait over 10s target

A 30-second dictation defines wait time: wait = 30 ÷ × realtime. Quality is a hard constraint whenever a model stays below 20% on every language; time targets relax first. Experimental rows have no sub-20% model but remain below 30%, or have a 95% confidence interval whose lower bound reaches 30%. Unbenchmarked and conclusively 30%+ pairs are omitted.