whatmodel.sh / stt / Granite Speech 3.3 2B

Granite Speech 3.3 2B

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

Granite Speech 3.3 2B is an open-weights speech-to-text model from IBM, released June 2025. Its published size is 3.01B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact checkpoint ibm-granite/granite-speech-3.3-2b (current revision 3.3.2; card 'Release Date: June 19, 2025'; repo first created 2025-04). Params: HF safetensors total. vLLM flags from the card (speech LoRA).

59.9±2.5#34 of 162 speech-to-text modelsfrom 3 benchmark results across 2 sources

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 Granite Speech 3.3 2B, out of 176 with a score, then more from IBM.

#modelscorewhereRAMreleased
39Qwen3-ASR-1.7B Alibaba (Qwen)60±1api $0.0005/audio min +local~3.3GB2026-01
40Canary-1B-Flash NVIDIA60±2local~4.4GB2025-03
41Granite Speech 3.3 2B IBM60±3local~7.7GB2025-06
42Speechmatics Enhanced Speechmatics60±2api $0.0067/audio min2023-03
43Deepgram Nova 2 Deepgram60±4api $0.0058/audio min2023-09
more from IBM
15Granite Speech 5.0 470M TurboCTC NC IBM64±4local~0.5GB2026-08
18Granite Speech 4.1 2B IBM62±3local~6GB2026-04
19Granite 4.0 1B Speech IBM62±3local~6GB2026-03

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

$ what it's good at

other tests3 not in a task · 98% of its score
testresultvs the others in its rundifficultycountswhy
AfriSpeech-MultiBench · 7-region African-accent mean WER25.5% ✎▂····▂···▂▆█ beats 4/11avg 46%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 41%typical
AfriSpeech-MultiBench · Parliament (legal domain) WER20.7% ✎▂▂····▂·▅·▃█ beats 7/12avg 52%▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 35%sharp
Open ASR English short-form WER (March 2026)6.0% ✎▁▁▁▃··▁▁▂██▆ beats 79/86avg 8.2%▇▇▇▇▇▇▇▇▇ 22%blunt

$ where to run it

Other ways to run it

  1. Run it as published · needs ~7.72 GB, NVIDIA only · from huggingface.co/ibm-granite
    vllm serve ibm-granite/granite-speech-3.3-2b --max-model-len 2048 --enable-lora --lora-modules speech=ibm-granite/granite-speech-3.3-2b --max-lora-rank 64

Weights and documentation: huggingface.co/ibm-granite ↗

$ how the score was worked out · AfriSpeech-MultiBench (arXiv:… 76%, Open ASR Leaderboard 22%

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
AfriSpeech-MultiBench (arXiv:2511.14255v1) · African-accented English ASR benchmark suite2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 76%CC BY 4.0
Open ASR Leaderboard1▇▇▇▇▇▇▇ 22%Apache-2.0
prior · pull toward the middle▇ 2%
weighted average= 59.9

$ 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