whatmodel.sh / stt / AuroraASR-1M

AuroraASR-1M

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

AuroraASR-1M is an open-weights speech-to-text model from AuroraAI Research, released October 2026. Its published size is 0.000949073B parameters. Licence: apache-2.0.

Catalogue identity notes

Pinned public FP32 checkpoint and maker-documented Transformers runtime; custom code required. Weight count is 949,073 per architecture section; runtime memory is estimated, not measured.

Model information sources

Release date ↗ · Parameter count ↗

No comparable score yet. Its published results are below, but they don't overlap enough with other models to calculate a score.

$ how it compares

Scored speech-to-text models released closest to AuroraASR-1M, out of 178 with a score.

#modelscorewhereRAMreleased
1Muse Voice Transcribe 1.0 Meta71±3api $0.0030/audio min2026-09
2Grok Voice Transcribe 2.0 xAI70±3api $0.0017/audio min2026-09
4Universal-3.6 Pro Realtime AssemblyAI70±3api $0.0075/audio min2026-09
5Linden 1 (Agent STT) Speechmatics69±3api $0.0050/audio min2026-09
24Qwen-Audio-3.0-ASR Alibaba62±4comparison only2026-09

see all speech-to-text models · find the ones that fit your machine

$ what it's good at

other tests2 not in a task · 0.0% of its score
testresultvs the others in its rundifficultycountswhy
LibriSpeech test-clean WER lab11.4% ✎▁·▃·▁▁▁▁▃▇█▆ beats 2/102—not counted yet: no other model in the same run to compare it with
LibriSpeech test-other WER lab26.5% ✎▃···▁▁▁▆▆█▆▁ beats 0/56—not counted yet: no other model in the same run to compare it with

$ where to run it

Other ways to run it

  1. Run it as published · needs ~0.5 GB · from huggingface.co/AuroraAI-Research (official)
    python -m pip install torch transformers sentencepiece soundfile && python -c "import soundfile as sf; from transformers import AutoModelForCTC; model = AutoModelForCTC.from_pretrained('AuroraAI-Research/AuroraASR-1M', trust_remote_code=True).eval(); wav, sr = sf.read('audio.wav', dtype='float32'); print(model.transcribe(wav)[0])"

Official weights and documentation: huggingface.co/AuroraAI-Research ↗

$ published results · 2 results, not enough overlap for a score
testresultsource and conditions
LibriSpeech test-clean WER11.4%AuroraASR-1M model card · lab-reported

FP32, greedy CTC, no external language model; full LibriSpeech test sets.

LibriSpeech test-other WER26.5%AuroraASR-1M model card · lab-reported

FP32, greedy CTC, no external language model; full LibriSpeech test sets.

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json