Cohere Transcribe needs about 2.2 GB of memory at Q4_K_M (a 1.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 6 quants, from 1.6 GB (Q4_K_M) to 4.1 GB (F16).
Cohere Transcribe: 2B‑parameter offline speech‑to‑text model
Cohere Transcribe is a 2‑billion‑parameter ASR model that converts 16 kHz mono WAV files into text. It uses a Conformer encoder and a Transformer encoder‑decoder head with cross‑attention. The model supports 14 languages and is available in several GGUF quantizations.
Good at
- Accurate multilingual transcription
- Low‑latency inference with GGUF
- Supports 14 languages
Limits
- No vision or image support
- Requires 16 kHz mono WAV input
- License is Apache‑2.0
Best for
- Transcribing podcasts
- Generating subtitles for videos
- Voice‑to‑text note taking
Who it’s for. Developers building offline speech‑to‑text pipelines who need a lightweight, open‑source model. The model is suitable for edge devices and local inference where internet access is limited.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 1.6 GB | 2.2 GB | Fits | Get file → | |
| Q5_K_M | 1.8 GB | 2.5 GB | Fits | Get file → | |
| Q6_K | 2.0 GB | 2.7 GB | Fits | Get file → | |
| Q8_0 | 2.4 GB | 3.2 GB | Fits | Get file → | |
| BF16 | 4.1 GB | 5.1 GB | Fits | Get file → | |
| F16 | 4.1 GB | 5.1 GB | Fits | Get file → |
For 8 GB VRAM, Q4_K_M (1.56 GB) or Q5_K_M (1.77 GB) are suitable, requiring about 2.2–2.5 GB memory at 8 k context. If you have 12 GB VRAM, Q6_K (1.97 GB) or Q8_0 (2.41 GB) can be used, needing roughly 2.7–3.2 GB memory. BF16 and F16 are larger and need more than 5 GB memory.
What GPU do you need for Cohere Transcribe?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q8_0 (2.4 GB file, about 3.2 GB in use).
- 6 GB VRAM: the best quant that fits is F16 (4.1 GB file, about 5.1 GB in use).
Which Cohere Transcribe quant should you choose?
Q4_K_M is the usual starting point for Cohere Transcribe: it balances file size and quality well. Here it is a 1.6 GB file.
The Q8_0 file is 2.4 GB, about 1.5 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Cohere Transcribe
How much VRAM does Cohere Transcribe need?
About 2.2 GB at Q4_K_M with an 8K context window. That is the 1.6 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Cohere Transcribe on an 8 GB GPU?
Yes, with F16: the 4.1 GB file needs about 5.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Cohere Transcribe GGUF?
It depends on the quant: Q4_K_M is 1.6 GB, Q5_K_M is 1.8 GB, Q6_K is 2.0 GB, Q8_0 is 2.4 GB, BF16 is 4.1 GB, F16 is 4.1 GB.
Where can I download Cohere Transcribe GGUF?
The files are on Hugging Face in the handy-computer/cohere-transcribe-03-2026-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Cohere Transcribe use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Cohere Transcribe on 8 GB VRAM?
Yes, using Q4_K_M or Q5_K_M quantizations will fit within 8 GB VRAM with about 2.2–2.5 GB memory usage.
Does the model support languages other than English?
Yes, it supports 14 languages including French, German, Spanish, Italian, Portuguese, Dutch, Polish, Greek, Arabic, Japanese, Chinese, Vietnamese, and Korean.
What license applies to this model?
The model is released under the Apache‑2.0 license, inherited from the parent model.