Whisper VRAM requirements
OpenAI's open speech-recognition models in five sizes from tiny to large, plus turbo, an optimized version of large-v3 that transcribes faster. OpenAI's figures run from about 1 GB for base (74M) and tiny (39M) to about 10 GB for large (1,550M parameters). The cheapest single GPU with 10 GB or more holds all of them: the 16 GB Tesla V100 at Verda (DataCrunch), $0.26 an hour.
What OpenAI says each setting needs
| Model and setting | Stated GPU memory | Cheapest single GPU with that much | Per hour | |
|---|---|---|---|---|
| large (1,550M parameters)Multilingual; the relative speed baseline | about 10 GB | Tesla V100, 16 GB, Verda (DataCrunch) | $0.26 | Rent → |
| turbo (809M)Multilingual; about 8x the speed of large | about 6 GB | Tesla V100, 16 GB, Verda (DataCrunch) | $0.26 | Rent → |
| medium (769M)English-only or multilingual | about 5 GB | Tesla V100, 16 GB, Verda (DataCrunch) | $0.26 | Rent → |
| small (244M)English-only or multilingual | about 2 GB | Tesla V100, 16 GB, Verda (DataCrunch) | $0.26 | Rent → |
| base (74M) and tiny (39M)English-only or multilingual | about 1 GB | Tesla V100, 16 GB, Verda (DataCrunch) | $0.26 | Rent → |
The memory figures are OpenAI's, not our estimate. The price is the lowest on-demand rate for one GPU with at least that much memory, from our latest check; a card that holds a model is not a promise of its speed.
Before you rent
OpenAI calls these approximate memory requirements. Even large needs only about 10 GB, so the cheapest card with 10 GB or more runs every size and the choice comes down to speed and price. One catch before you pick turbo: OpenAI says it is not trained for translation, so to translate speech into English use large or another multilingual size.
In OpenAI's words
- large (1,550M parameters): “| large | 1550 M | N/A | `large` | ~10 GB | 1x |”
- turbo (809M): “| turbo | 809 M | N/A | `turbo` | ~6 GB | ~8x |”
- medium (769M): “| medium | 769 M | `medium.en` | `medium` | ~5 GB | ~2x |”
- small (244M): “| small | 244 M | `small.en` | `small` | ~2 GB | ~4x |”
- base (74M) and tiny (39M): “| base | 74 M | `base.en` | `base` | ~1 GB | ~7x |”
Source: the Whisper README on GitHub, read 17 October 2026.
Text models are sized by a formula instead: LLM VRAM requirements for every open language model, or the LLM VRAM calculator for your own context and batch.
More models sized by their publishers' figures
- Stable Diffusion 3.5 Medium (image generation, Stability AI)
- Wan2.1 (video generation, Wan-AI)
- Wan2.2 (video generation, Wan-AI)
- HunyuanVideo (video generation, Tencent)
- olmOCR (OCR and PDF conversion, Ai2)
- MinerU (OCR and document parsing, OpenDataLab)
- GTE-Qwen2 (text embeddings, Alibaba)