Default · 1.6 GB
Whisper large-v3-turbo
Fast and multilingual — 99 languages with the best all-round balance of speed and accuracy. What most people should use, so it’s what ships selected.
Feature
The default speech model — Whisper large-v3-turbo — covers 99 languages including Hindi, Indian English, Mandarin, Arabic, Spanish, and French. Set a language, or let Velora detect it mid-sentence.
The model menu
Every option downloads once and runs on your Mac. Velora defaults to the best all-rounder and lets you trade accuracy, speed and disk when your needs are sharper.
Default · 1.6 GB
Fast and multilingual — 99 languages with the best all-round balance of speed and accuracy. What most people should use, so it’s what ships selected.
Accuracy · 3.1 GB
The full-size model: same languages, more headroom on hard accents and noisy rooms. Slower and larger — a deliberate trade.
Hinglish · 1.6 GB
Fine-tuned on 700+ hours of Hindi–English code-switching, with Romanized output. Built for the way Hinglish is actually spoken, not textbook Hindi.
Speed · 2.3–2.5 GB
The lowest-latency finals: v2 for pure English, v3 adding 24 European languages. No Hindi, Mandarin, or Arabic — pure speed for the languages they cover.
Small · 0.5 GB
The smallest multilingual option for tight disks — roughest quality, still 99 languages, still local.
Experimental · 0.9 GB
A faster multilingual Whisper turbo through a different runtime. Marked experimental because it is newer and less proven than the default — try it, and switch back if it lets you down.
Cleanup tiers
Qwen3.5 in three sizes: 2B 4-bit on Macs under 14 GB, 4B 4-bit from 14 GB, and 4B 8-bit from 24 GB. Velora picks the tier that fits your RAM; you can override it.
Free. Open source. Local.
Already using Velora? A star is how the next Mac user finds it.