How much VRAM does cohere-transcribe-03-2026 need?

Q4_K_M is exactly 1,558,162,944 bytes (1.45 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

CohereLabs · audio asr

cohere-transcribe-03-2026

CohereLabs/cohere-transcribe-03-2026

cohere-transcribe-03-2026 at Q4_K_M is exactly 1,558,162,944 bytes (1.45 GiB / 1.56 GB) — an effective 6.034 bits per weight, not the nominal 4.

From the file· summed from 1 file(s)

Parameters

2.1B

Architecture

cohere_asr

Context

—

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q4_K	1.41 GiB	1,510,363,008	5.849	—	cstr
Q4_K_M	1.45 GiB	1,558,162,944	6.034	2103	handy-computer
Q5_0	1.62 GiB	1,738,723,200	6.733	—	cstr
Q5_K_M	1.65 GiB	1,770,270,208	6.856	2103	handy-computer
Q5_1	1.73 GiB	1,852,903,296	7.176	—	cstr
Q6_K	1.84 GiB	1,972,524,544	7.639	2103	handy-computer
Q6_K	1.85 GiB	1,981,355,904	7.673	2104	cstr
Q8_0	2.25 GiB	2,410,655,232	9.335	2103	handy-computer
Q8_0	2.26 GiB	2,423,803,776	9.386	2104	cstr
BF16	3.82 GiB	4,105,263,104	15.898	—	handy-computer
F16	3.82 GiB	4,106,644,992	15.903	2103	handy-computer

Measured

published by a third party, attributed below

Metric	Value	What it means
rtf	915.6
RTFx	915.6	Higher is better — audio seconds processed per second of compute.
Word error rate	0.96%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.88%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.80%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.93%	Lower is better — the share of words transcribed incorrectly.
Word error rate	5.58%	Lower is better — the share of words transcribed incorrectly.
Word error rate	5.38%	Lower is better — the share of words transcribed incorrectly.
Word error rate	2.74%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.01%	Lower is better — the share of words transcribed incorrectly.
Word error rate	10.38%	Lower is better — the share of words transcribed incorrectly.
Word error rate	2.04%	Lower is better — the share of words transcribed incorrectly.
Word error rate	5.20%	Lower is better — the share of words transcribed incorrectly.

Benchmarked· by open-asr-leaderboard-english-short-latest

RTFx measured by the Open ASR Leaderboard on a single datacenter GPU at a large batch size. It ranks models against each other; it says nothing about throughput on consumer hardware. We reproduce these figures with attribution; they are not ours and we have not verified the runs. Source: open-asr-leaderboard-english-short-latest.

Compare with

same modality, comparable size

vs nemotron-3.5-asr-streaming-0.6b vs parakeet-unified-en-0.6b vs parakeet-tdt-0.6b-v3 vs whisper-medium vs Voxtral-Mini-4B-Realtime-2602 vs whisper-large-v3 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 1.08 GiB. The real file is 1.45 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

Architecture unavailable — this repository is gated and no ungated mirror was found. Exact file sizes above are still authoritative; only the KV math needs the config.

Questions people ask

How much VRAM does cohere-transcribe-03-2026 need?: Q4_K_M is exactly 1,558,162,944 bytes (1.45 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of cohere-transcribe-03-2026 should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.