Whisper models

30 Whisper models from OpenAI on Hugging Face, from 242M to 1.6B parameters, published by OpenAI in F32 and F16. The smallest official model, whisper-small.en, needs about 1.1 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.

Pick a size

One row per official Whisper size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.

ModelParametersNative VRAMFP8 VRAMINT4 VRAMLive GPU fit (native)Est. $/hrKV cache at 32K
whisper-small242M1.1 GB0.3 GB0.1 GBRTX 3070$0.088/hrnot published for this architecture
whisper-large-v3-turbo809M1.8 GB0.9 GB0.5 GBRTX 3070$0.088/hrnot published for this architecture
whisper-large-v31.5B3.4 GB1.7 GB0.9 GBRTX 3070$0.088/hrnot published for this architecture

VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.

Official models (8)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
whisper-small.en242MF321.1 GBRTX 3070$0.088/hr
whisper-small242MF321.1 GBRTX 3070$0.088/hr
whisper-medium.en764MF323.4 GBRTX 3070$0.088/hr
whisper-medium764MF323.4 GBRTX 3070$0.088/hr
whisper-large-v3-turbo809MF161.8 GBRTX 3070$0.088/hr
whisper-large1.5BF326.9 GBRTX 3070$0.088/hr
whisper-large-v21.5BF326.9 GBRTX 3070$0.088/hr
whisper-large-v31.5BF163.4 GBRTX 3070$0.088/hr

Fine-tunes and community models (22)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
whisper-bemba-stt242MF321.1 GBRTX 3070$0.088/hr
kotoba-whisper-bilingual-v1.0756MBF161.7 GBRTX 3070$0.088/hr
kotoba-whisper-v2.0756MBF161.7 GBRTX 3070$0.088/hr
kotoba-whisper-v2.2756MF323.4 GBRTX 3070$0.088/hr
anime-whisper756MF323.4 GBRTX 3070$0.088/hr
tugstugi_bengaliai-regional-asr_whisper-medium764MF323.4 GBRTX 3070$0.088/hr
nb-whisper-medium764MF323.4 GBRTX 3070$0.088/hr
whisper-large-v3-turbo-armenian809MF323.6 GBRTX 3070$0.088/hr
whisper-large-v3-turbo809MF323.6 GBRTX 3070$0.088/hr
Whisper-Hindi2Hinglish-Apex809MBF161.8 GBRTX 3070$0.088/hr
whisper-large-v3-turbo-german809MBF161.8 GBRTX 3070$0.088/hr
kazakh-whisper-large-v3-turbo809MF161.8 GBRTX 3070$0.088/hr
Breeze-ASR-251.5BBF163.4 GBRTX 3070$0.088/hr
Breeze-ASR-261.5BF326.9 GBRTX 3070$0.088/hr
whisper-large-v3-russian1.5BBF163.4 GBRTX 3070$0.088/hr
whisper-large-v3-ru-podlodka1.5BF326.9 GBRTX 3070$0.088/hr
whisper-large-v31.5BF326.9 GBRTX 3070$0.088/hr
nb-whisper-large1.5BF326.9 GBRTX 3070$0.088/hr
Whisper-Hindi2Hinglish-Prime1.5BF326.9 GBRTX 3070$0.088/hr
whisper-large-v3-german1.5BBF163.4 GBRTX 3070$0.088/hr
kb-whisper-large1.6BF163.6 GBRTX 3070$0.088/hr
CrisperWhisper1.6BF163.6 GBRTX 3070$0.088/hr

VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.

More from OpenAI

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.