Whisper large-v3
Release in the Whisper family · version large-v3
Whisper large-v3 is a 1550M-parameter multilingual speech recognition and translation model released by OpenAI in November 2023. It keeps the architecture of the earlier large models but uses 128 Mel frequency bins instead of 80 and adds a language token for Cantonese.314
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Repository: openai/whisper repository (external site: github.com)
- Release notes: large-v3 release discussion (external site: github.com)
- License: LICENSE (MIT, repository) (external site: github.com)
Availability and license
Overall availability
Downloadable without gating from Hugging Face, and fetched automatically by the openai-whisper package when "large-v3" (or its alias "large") is loaded.273
Availability is separate from permission: read the license before using or redistributing.
MIT License (openai/whisper repository) (external site: github.com)65
Apache License 2.0 (Hugging Face model card metadata) (external site: huggingface.co)12
The sources disagree. The openai/whisper README states that Whisper's code and model weights are released under the MIT License, while the Hugging Face repository for large-v3 declares Apache-2.0 in its card metadata. Both are permissive licenses; this catalog records both rather than choosing one.51
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Published on Hugging Face without gating and downloadable through the openai-whisper package.27 |
| Inference codeIs code for running the model published? | Public | The openai/whisper repository provides the inference code and command-line tool; the Hugging Face card documents inference with Transformers.51 |
| Training codeIs the code used to train the model published? | Unknown | The repository covers inference and evaluation-data preparation; this review found no published training code.5 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | Trained on 1 million hours of weakly labeled audio and 4 million hours of audio pseudo-labeled with large-v2. The data is described but not released.31 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | OpenAI states the model was trained for 2.0 epochs over this mixture and lists the architectural changes from large-v2; the original paper describes the general training approach. No training configuration is published.39 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The README and release discussion report word and character error rates for large-v3 on Common Voice 15 and FLEURS. The repository documents how the paper's evaluation datasets were prepared, but no large-v3 evaluation scripts were located.538 |
What it is useful for
Multilingual transcription and translation of speech into English. The model card names AI researchers as the primary intended users, says the models may also serve developers as a speech recognition solution, and advises against non-consensual transcription, classifying people, and high-risk decision-making uses.41
Run and use notes
Organization context
Provenance and derivatives
Trained by OpenAI. Part of its training data was pseudo-labeled with OpenAI's earlier Whisper large-v2 model.31
- Derived from: Whisper large-v2 (pseudo-labels) (external site: huggingface.co) — Used to pseudo-label 4 million hours of training audio; no separate catalog record.
Catalog records that name this entry in their provenance:
Other releases in the Whisper family
- Whisper large-v3-turboModel-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.