Whisper large-v3-turbo

OpenAI's MIT-licensed speech-to-text model: the turbo version is about 8x faster than large and runs in roughly 6 GB of GPU memory.

In short. Whisper turns spoken audio into written text. It is one of the most widely used open speech recognition models, it handles many languages and accents, and it can also translate speech into English. The turbo version is a faster, slimmer edition that fits on an ordinary gaming GPU. The license is MIT.

Summary

Whisper is OpenAI's open speech recognition model, first released in 2022. Whisper large-v3-turbo, announced on 1 October 2024, is a pruned and fine-tuned version of large-v3 in which the number of decoder layers drops from 32 to 4. That makes it much faster at the cost of a minor quality loss. OpenAI's repository lists it at 809 million parameters, about 6 GB of VRAM and roughly 8 times the speed of the large model, which has 1,550 million parameters and needs about 10 GB. OpenAI measured the speeds on English speech on an A100, so real-world results vary.

Whisper transcribes speech in the same language it hears, and the larger multilingual checkpoints can also translate speech into English. The turbo model is not trained for translation: it returns the original language even when asked to translate, so use large-v3 or another multilingual size for that. Whisper can return sentence-level or word-level timestamps. It looks at 30-second windows, and long recordings are handled by a sliding-window or chunked algorithm. The large-v3 checkpoint was trained on 1 million hours of weakly labeled audio and 4 million hours of pseudo-labeled audio.

Whisper's code and weights are published under the MIT license. OpenAI's card says it shows strong results in about ten languages, performs unevenly across languages, accents and dialects, and can hallucinate text that was never spoken or repeat itself. It cannot do real-time transcription out of the box. Newer open models such as NVIDIA's Nemotron 3.5 Transcribe now compete with it, but Whisper remains a common baseline with a large ecosystem of tools around it.

Good for

  • Transcribing meetings, interviews, podcasts and videos on your own computer
  • Adding subtitles with sentence-level or word-level timestamps
  • Translating non-English speech into English with the large-v3 checkpoint
  • A well-supported baseline with many community runtimes and apps

Watch out for

  • It can invent words that were never spoken, especially in silence or noise, so check important transcripts
  • Accuracy varies a lot by language, accent and dialect; test your own audio
  • The turbo model does not translate; use large-v3 for speech translation
  • Hugging Face labels the large-v3 page Apache 2.0, while the GitHub repository and the turbo page say MIT; both are permissive, so keep the license text for the checkpoint you use

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of Whisper large-v3-turbo. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.

Variants Parameters Download size Memory needed Fits on
large-v3-turbo FP16 810M 1.6 GB 6 GB Laptop or phone
large-v3 FP16 1.54B 3.1 GB 10 GB Laptop or phone

Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 6 GB of GPU memory.

The license in plain words

The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.

More about MIT

See also

Frequently asked questions

What is Whisper large-v3-turbo?
Whisper turns spoken audio into written text. It is one of the most widely used open speech recognition models, it handles many languages and accents, and it can also translate speech into English. The turbo version is a faster, slimmer edition that fits on an ordinary gaming GPU. The license is MIT.
Can I use Whisper large-v3-turbo commercially?
Yes. Whisper large-v3-turbo is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is Whisper large-v3-turbo open source?
Whisper large-v3-turbo is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run Whisper large-v3-turbo on my own computer?
Yes. The smallest variant, large-v3-turbo, needs about 6 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was Whisper large-v3-turbo released?
Whisper large-v3-turbo was released by OpenAI on October 1, 2024.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Parakeet TDT 0.6B v3 NVIDIA 627M – CC BY 4.0 (Permissive) 2.89 GB Laptop or phone
Cohere Transcribe Cohere 2.07B – Apache 2.0 (Permissive) 4.75 GB Laptop or phone
Nemotron 3.5 Transcribe NVIDIA 34B 4.3B active 64K OpenMDW-1.1 (Permissive) 25 GB Gaming GPU
Qwen3-ASR Alibaba Qwen 918M – 2.35B – Apache 2.0 (Permissive) 2.12 GB Laptop or phone