whatmodel.sh / stt / GigaAM Multilingual Large (CTC 600M)

GigaAM Multilingual Large (CTC 600M)

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

GigaAM Multilingual Large (CTC 600M) is an open-weights speech-to-text model from Sber, released July 2026. Its published size is 0.6B parameters. Licence: mit.

Catalogue identity notes

ai-sage/GigaAM-Multilingual branch 'large_ctc' @3905cd5 (SaluteDevices/Sber GigaAM team; MIT; repo created 2026-07-14): 600M Conformer ASR model with a character-wise CTC decoder fine-tuned on 50K hours (Russian, Kazakh, Kyrgyz, Uzbek best; 'moderate quality on English'). Distinct from the ssl/large_ssl encoders on other branches of the same repo and from the earlier Russian-only GigaAM v2/v3 releases.

Model information sources

Release date ↗ · Parameter count ↗

Card: 'a family of Conformer-based foundation models (220M / 600M parameters)'; 'large_ctc' variant.

46.5±7.0#114 of 162 speech-to-text modelsfrom 2 benchmark results across 1 source* thinly tested, treat the rank as provisional

Score within the Open ASR English short-form WER (March 2026) comparison set only.

$ how it compares

The speech-to-text models ranked either side of GigaAM Multilingual Large (CTC 600M), out of 176 with a score, then more from Sber.

#modelscorewhereRAMreleased
125VibeVoice-ASR-Streaming-7B Microsoft47±5local~17GB2026-09
126SpeechBrain Conformer (LargeScaleASR / Loquacious) SpeechBrain47±5local~2.8GB2025-02
127GigaAM Multilingual Large (CTC 600M) Sber47±7local~2.3GB2026-07
128STT En Conformer-Transducer Large NVIDIA46±4local~1.1GB2022-09
129ESPnet E-Branchformer (no LM) ESPnet46±4local~0.8GB2023-01
more from Sber
147GigaAM Multilingual (CTC 220M) Sber42±7local~0.9GB2026-07

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

$ what it's good at

other tests2 not in a task · 81% of its score
testresultvs the others in its rundifficultycountswhy
Common Voice English WER (version unspecified) lab21.5% ✎█···█·█····█ beats 1/3avg 20%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 61%typical
FLEURS English test WER lab9.4% ✎▁··▁▁▁▃▂▃█▂▁ beats 8/64avg 6.6%▇▇▇▇▇▇▇▇ 20%blunt

$ where to run it

Other ways to run it

  1. Run it as published · needs 2.34 GB · from huggingface.co/ai-sage
    pip install "transformers>=5" torch torchaudio hydra-core omegaconf && python -c "from transformers import AutoModel; model = AutoModel.from_pretrained('ai-sage/GigaAM-Multilingual', revision='large_ctc', trust_remote_code=True); print(model.transcribe('example.wav'))"

Weights and documentation: huggingface.co/ai-sage ↗

$ how the score was worked out · ai-sage/GigaAM-Multilingual m… 81%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
ai-sage/GigaAM-Multilingual model card (Model Performance table) · lab-reported2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 81%MIT
prior · pull toward the middle▇▇▇▇▇▇ 19%
weighted average= 46.5

$ 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