Whisper Tiny is an Apache-2.0 licensed automatic speech recognition model from OpenAI, available as openai/whisper-tiny with safetensors tags. It is a Transformer-based encoder-decoder (sequence-to-sequence) model trained on 680k hours of labelled speech using large-scale weak supervision. The model supports English and multilingual speech recognition and speech translation. It has 39M parameters and is part of the Whisper family of five configurations. Reported LibriSpeech test WER is 7.54 on clean and 17.15 on other for English, and 141 on Common Voice 11.0 Hindi test.
Model size
0.04B
Format
FP32 weights
Freshness 0 out of 100. Repository age: 1,480 days old, Sep 26, 2022. Measures age, not quality.Freshness0/1001,480 days old
Popularity 79 out of 100. 892,741 downloads in 30 days; 467 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity79/100892.7K downloads / 30d
Whisper base is a pre-trained Transformer encoder-decoder model for automatic speech recognition and speech translation. It was trained on 680k hours of labelled speech data using large-scale weak supervision by OpenAI. The model is available under the Apache-2.0 license and uses the safetensors format. It is a multilingual model, supporting many languages as listed on its model card. Reported metrics include a test WER of 5.008769117619326 on LibriSpeech clean, 12.84936273212057 on LibriSpeech other, and 131 on Common Voice 11.0 Hindi. The model card states it can generalise to many datasets and domains without fine-tuning. It comes in five configurations of varying model sizes, and this base checkpoint is one of them.
Model size
0.07B
Format
FP32 weights
Freshness 0 out of 100. Repository age: 1,480 days old, Sep 26, 2022. Measures age, not quality.Freshness0/1001,480 days old
Popularity 81 out of 100. 1,628,745 downloads in 30 days; 306 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity81/1001.6M downloads / 30d
whisper-small is an Apache-2.0 licensed automatic speech recognition (ASR) model from openai/whisper-small, stored in safetensors format. It is a Transformer-based encoder-decoder (sequence-to-sequence) model trained on 680k hours of labeled speech data using large-scale weak supervision. According to the official model card, it supports multilingual speech recognition and speech translation; for ASR it predicts transcriptions in the same language as the audio, and for translation it predicts transcriptions in a different language. The model card lists many supported languages, including English, Chinese, German, Spanish, Russian, Korean, French, Japanese, Portuguese, Turkish, Polish, Catalan, Dutch, Arabic, and others. Reported model-index results include Test WER 3.432213777886737 on LibriSpeech (clean), Test WER 7.628304527060248 on LibriSpeech (other), Test WER 87.3 on Common Voice 11.0 (hi), and Test WER 125.69809089960707 on Common Voice 13.0 (dv).
Model size
0.24B
Format
FP32 weights
Freshness 0 out of 100. Repository age: 1,480 days old, Sep 26, 2022. Measures age, not quality.Freshness0/1001,480 days old
Popularity 86 out of 100. 3,373,098 downloads in 30 days; 629 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity86/1003.4M downloads / 30d
Whisper Medium is a pre-trained Transformer-based encoder-decoder model for automatic speech recognition and speech translation. It is trained on 680k hours of labelled speech data using large-scale weak supervision, supports a large set of languages, and is available under the Apache 2.0 license. The model card reports WER of 2.9 on LibriSpeech clean, 5.9 on LibriSpeech other, and 53.87 on Common Voice 11.0 Hindi test.
Model size
0.76B
Format
FP32 weights
Freshness 0 out of 100. Repository age: 1,480 days old, Sep 26, 2022. Measures age, not quality.Freshness0/1001,480 days old
Popularity 73 out of 100. 303,535 downloads in 30 days; 313 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity73/100303.5K downloads / 30d
Open-weight multilingual speech recognition model for transcribing local recordings. The model card includes weights and a Transformers inference path.
Model size
809M
Freshness 6 out of 100. Repository age: 744 days old, Oct 1, 2024. Measures age, not quality.Freshness6/100744 days old
Popularity 92 out of 100. 6,349,781 downloads in 30 days; 3,454 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity92/1006.3M downloads / 30d
Whisper large is a pre-trained Transformer-based encoder-decoder model for automatic speech recognition and speech translation, developed by OpenAI. It was trained on 680k hours of labeled speech data using large-scale weak supervision and is released under the Apache-2.0 license. The official model card notes that the large-v2 variant is recommended in place of this original large model, as it was trained for 2.5x more epochs with regularization and surpasses its performance without architecture changes.
Model size
1.54B
Format
FP32 weights
Freshness 0 out of 100. Repository age: 1,480 days old, Sep 26, 2022. Measures age, not quality.Freshness0/1001,480 days old
Popularity 61 out of 100. 22,439 downloads in 30 days; 564 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity61/10022.4K downloads / 30d