by OpenAI
Whisper Base.
audio
open weights
edge
0B params
30 ctx
🧬 Distilled from
Whisper Large v3
— smaller, cheaper to run, similar reasoning style.
Smallest GPU
1× Nvidia Titan V
74M Whisper — browser / Raspberry Pi-deployable.
Smallest GPU to run it
See all quantisations →
1× Nvidia Titan V.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Where to use it
Cheapest hosted endpoints.
| Provider | Access | $/M in | $/M out | |
|---|---|---|---|---|
| OpenAI API | api direct | — | — | Launch ↗ |
Family
Whisper Large v3
Whisper Medium
Whisper Small
Whisper Tiny
Variants in the Whisper family.
Whisper Large v3
1.6B
open
MIT
OpenAI's open-weight speech-to-text — the standard transcription model.
Whisper Medium
0.8B
open
MIT
769M Whisper variant — half the size of Large, 80% of the accuracy.
Whisper Small
0.2B
open
MIT
244M Whisper — fits on edge GPUs and CPU.
Whisper Tiny
0.0B
open
MIT
39M Whisper — runs in-browser via WebGPU.
FAQ
Frequently asked.
How do I run Whisper Base?
Whisper Base is open-weight, so you can self-host on rented GPUs. See the Run It Yourself tab for GPU configurations + cost estimates, or use one of the hosted inference providers listed on this page.
Where can I access Whisper Base?
Whisper Base is available via OpenAI API. Each access option lists its own pricing (per million tokens or hourly hosting).
How much does it cost to run Whisper Base?
Self-hosting cost depends on the GPU you rent and the throughput you need. See the Run It Yourself tab for GPU configurations and hourly cost estimates.
Is Whisper Base open-source or proprietary?
Whisper Base is open-weight under the MIT license. You can download and self-host it.
Run it yourself
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr refreshed hourly from each provider's API.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
3 GB | — | — | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
3 GB | — | — | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
3 GB | — | — | Compare → |
Just need an API?
Skip the GPU rental and call a hosted endpoint instead.
See access providers on the Overview tab →
Context window
How much it can remember.
30 tokens
≈ 23 English words
4K
32K
128K
1M
Capabilities
What it can do.
·
Vision input
·
Audio input
·
Video input
·
Function calling
·
Tool use
·
JSON mode
✓
Streaming
·
Fine-tuning