Whisper · speech to text
Whisper medium
OpenAI's README lists 769 M for this size. The exact count, 763,857,920, is Computed: I added up the tensor shapes in the official model.safetensors header on 19 Nov 2026.
OpenAI's GitHub README says the code and model weights are released under MIT. The Hugging Face repo metadata says apache-2.0 for every size except large-v3-turbo (mit). I follow the GitHub statement and flag the mismatch.
What it is
Whisper writes down speech. It is not a chat model, so this page has no context length and no KV memory math. OpenAI’s name for this file is medium. The README lists 769 M parameters. The exact count, 763,857,920, is Computed from the official safetensors header on 19 Nov 2026.
99 language codes (Computed: tokenizer.py lists 100, and Cantonese only arrived with large-v3). OpenAI's card says strong results in about 10 languages.
From the model card
Strengths
- OpenAI suggests medium or large for translating into English.
- Aiko uses it on iPhones with enough memory.
Weaknesses
- Not in Argmax's iPhone list. Better on a Mac or PC.
File size and memory
These are not the Frankenstein run. Each row keeps the label it came with.
| Figure | Value | Label |
|---|---|---|
| OpenAI 'Required VRAM' (openai-whisper, PyTorch) | ~5 GB | Vendor-reported Source |
| OpenAI relative speed vs large (A100, English) | ~2x | Vendor-reported Source |
| Hugging Face model.safetensors (float32) | 3.06 GB (3,055,544,304 bytes) | Vendor-reported Source |
| FP16 weights (params x 2 bytes) | 1.42 GiB | Computed params from the safetensors header |
| whisper.cpp ggml file | 1.5 GiB | Vendor-reported Source |
| whisper.cpp runtime memory | ~2.1 GB | Vendor-reported Source |
Speed
| Model | Audio | File on disk | Wall time | Audio ÷ wall | Encoder | Peak RSS | JFK word errors |
|---|---|---|---|---|---|---|---|
| medium | JFK sample, 11.0 s | 1,463 MiB | 8.93 s | 1.23× | 4.36 s | 2,068 MiB | 0.0% (0/22 words) |
| medium | LibriVox Gettysburg Address, 161.2 s | 1,463 MiB | 60.64 s | 2.66× | 25.28 s | 2,171 MiB | — |
Method, commit, and the exact command: methodology.
Measured on Frankenstein (HP ProLiant DL380p Gen8), 19 Nov 2026. A phone speed is not in this run. The Tesla P40 was not used.
Where it runs
Vendor-reported Not listed for any iPhone chip. This is Argmax’s list for WhisperKit, not a test I ran, and it names chips rather than phone models. See Run Whisper on your iPhone and Run Whisper on your computer.
Official downloads
These go to OpenAI or to the whisper.cpp GGML files. They are not store links.
- openai-whisper (pip), model name: medium
- Hugging Face openai/whisper-medium
- whisper.cpp ggml-medium.bin
Sources: openai/whisper README (model table) · openai/whisper model card · openai/whisper-medium model card · whisper.cpp models list (ggml file sizes) · WhisperKit device support (Argmax config.json)