Granite Speech 4.1 2b NAR
IBM · released Mar 10, 2026 · ibm-granite/granite-speech-4.1-2b-nar
Speech to textTranscribes a recording into words
- Type
- Open weightsApache License 2.0
- Languages
- 5
- Size
- 2.3B
Context measured in tokens
Our take
Written Sep 5, 2026Granite Speech 4.1 is a 2.3-billion-parameter speech-to-text model from IBM that turns audio into written words under a permissive Apache licence. It is built for speed and for difficult audio conditions rather than polished studio recordings, running at over two thousand times real time on benchmark hardware.
Pick this for transcribing meeting-room or accented speech where it scores among the best measured, or when you need extreme batch speed on modest hardware. Use it for self-hosted European-language workloads across English, French, German, Spanish and Portuguese. Skip it if your audio is mostly podcasts or internet video, where it trails the field, or if you need a hosted provider rather than running it yourself.
The case for it
- Among the most accurate on recorded meeting-room audio: about seven words wrong per hundred, against a field middle of over ten.
- Better than most on accented speech, with a lower error rate than the typical model.
- Over two thousand times real time on benchmark hardware: an hour of audio in roughly two seconds.
- Apache 2.0 licence allows commercial use, fine-tuning and redistribution.
The case against it
- Worse than most on podcasts and internet video, with a higher error rate than the field middle.
- No hosted inference options listed; you must self-host.
How good is it?
An open speech-to-text model for turning recordings of meetings and clear read-aloud speech into written text.
- transcribing recordings of meetings in a roomRecorded meetings · 6th of 92
- transcribing clear recordings of people reading aloudClean read speech · 9th of 92
TranscriptionTurning speech into text4 of 5Open ASR WER · 23rd of 76
95.3%
Misses roughly one word in 21, averaged over nine English test sets.
2,074×26th of 74
an hour of audio in 2 seconds, on the board's own hardware. Your machine will differ.
5
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
Which languages ↓ ↑
English · French · German · Spanish · Portuguese
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 19.9 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 27.9 GB spare means a 10% error in the size would not change the answer.
Apple M2 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 3.1 GB spare means a 10% error in the size would not change the answer.
These cards answer whether Granite Speech 4.1 2b NAR loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- ibm-granite/granite-speech-4.1-2b-nar
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- ibm-granite-granite-speech-4-1-2b-nar