Short answer: on NVIDIA or Apple Silicon, start with Balanced from about 2 GB of graphics memory (Apple: unified memory); AMD and Intel graphics need about 4 GB for it. Stay on Quick & Compact below that or without a GPU. Best Quality is recommended from about 3.6 GB on NVIDIA, from 4 GB on Apple Silicon, and from 12 GB on AMD or Intel. All three tiers run fully offline, cover 99 languages, and you can switch at any time: the app re-downloads a single file. If the app language you pick during setup is one of the 25 European languages Parakeet supports, the setup assistant preselects Parakeet, which uses a GPU when one is available and otherwise runs on the processor, where it was several times faster than the compact Whisper model in our benchmark.
The three tiers, side by side
WhisPaste deliberately offers three tiers rather than the full whisper.cpp catalogue. Each is a quantised GGML build, which is why the downloads stay well under a gigabyte. You pick a tier in the settings; the app downloads one file, verifies its checksum and switches over.
| Tier | Model | Download | Graphics memory |
|---|---|---|---|
| Quick & Compact | whisper-small (q5_1) | 181 MB | ~900 MB |
| Balanced | whisper-medium (q5_0) | 514 MB | ~1.5 GB |
| Best Quality | whisper-large-v3-turbo (q5_0) | 547 MB | ~2.6 GB |
| Parakeet | parakeet-tdt-0.6b-v3 (int8) | 639 MB | GPU optional, else processor |
Get a recommendation
Your recommendation
JavaScript is off, so all tiers are described below. The comparison table above gives you the same information.
Quick & Compact — whisper-small
A 181 MB download that needs about 900 MB of graphics memory, and the tier that keeps a modest machine responsive. It is the right starting point on integrated graphics, on an older card, and on anything without a GPU at all. Accuracy is lower than the larger tiers on difficult audio; for clear speech into a decent microphone the difference is smaller.
Balanced — whisper-medium
A 514 MB download that wants about 1.5 GB of graphics memory. This is the tier most people should be on: usually more accurate than Quick & Compact on accents and technical vocabulary, without the memory appetite of the top tier. On Apple Silicon, 2 GB of unified memory is enough for it; from 4 GB the picker recommends Best Quality.
Best Quality — whisper-large-v3-turbo
A 547 MB download that wants about 2.6 GB of graphics memory. The turbo variant of large-v3 is the largest model WhisPaste ships. It is worth it with a dedicated card of about 4 GB or more, or with 4 GB or more of unified memory on Apple Silicon, and it is the tier to pick when your audio is hard: fast speech, unfamiliar names, specialist vocabulary.
Also consider: Parakeet
Parakeet is the second on-device engine, not a Whisper tier. The download is 639 MB for the processor bundle; with GPU acceleration enabled, the GPU model adds about 638 MB. On the processor it is still several times faster than Whisper — which makes it the strongest option on a machine without usable graphics acceleration. The trade-off is language coverage: 25 European languages instead of Whisper’s 99.
How the recommendation works
The picker above is not a guess — it applies the same thresholds the app uses. Dedicated NVIDIA cards are judged against roughly 70 % of their reported memory, because the operating system and other processes take their share: about 2.1 GB reported to reach Balanced, about 3.6 GB to reach Best Quality. Apple Silicon is treated generously, since unified memory is genuinely shared: 2 GB for Balanced, 4 GB for Best Quality. Integrated AMD and Intel graphics are treated the most cautiously of all — they report borrowed system memory as if it were their own, so Best Quality only gets recommended from about 12 GB upward.
Language decides more than hardware
Before optimising the tier, check the language. Whisper covers 99 languages at every tier, so a larger model does not unlock a new language — it only transcribes the same language more accurately. Parakeet is the opposite trade: much faster on a processor, restricted to 25 European languages including German, English, French, Spanish, Italian, Polish, Dutch, Ukrainian and Russian. If your language is on that list and your machine has no usable GPU, Parakeet will beat any Whisper tier on speed. The setup assistant preselects Parakeet when the app language you pick during setup is one of them.
Before you get started
- Which Whisper model size should I start with?
- Balanced (whisper-medium) from about 2 GB of graphics memory on NVIDIA or Apple Silicon, from 4 GB on AMD or Intel. Quick & Compact (whisper-small) below that or without a GPU. Switching later costs one download, so there is no need to agonise over it up front.
- Is the biggest model always the best choice?
- No. A model that does not fit your graphics memory can fail or run slowly. WhisPaste then falls back to a smaller model, so a model that fits comfortably is the safer choice. Accuracy also plateaus on clean audio from a decent microphone — the larger tiers earn their keep on accents, noise and specialist vocabulary.
- Can I change the model size later?
- Yes. The tier is a setting; the app downloads the matching file once, verifies it and uses it from then on. Nothing else in your setup changes.