Quantization format
Q6_K
Files labelled Q6_K average 6.808 effective bits per weight across 2999 real quantizations — not the nominal 6.5625. That is 4% more than the label implies, because a quantization is a mixture: some tensors are always kept at higher precision.
From the file· 2999 files measured
Nominal bpw
6.5625
from the block layout
Measured average
6.808
2999 files
Range
1.038–29.692
varies by architecture
File sizes
0.01 GiB+
up to 785.02 GiB
What files with this label actually contain
tensor types across 37 parsed files
F32
13560
Q6_K
11831
F16
999
Q8_0
778
MXFP4
72
BF16
39
Q4_0
16
If Q6_K were a uniform precision, this chart would have one bar. The F32 entries are normalization and bias tensors, which are never quantized; the higher K-quant entries are attention and output tensors deliberately promoted to protect quality.
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-27B | 20.57 GiB | 6.359 | +-3% |
| Qwen3.6-35B-A3B | 26.56 GiB | 6.345 | +-3% |
| Qwen3.5-9B | 7.16 GiB | 6.369 | +-3% |
| gemma-4-26B-A4B-it | 21.29 GiB | 6.890 | +5% |
| gemma-4-12B-it | 9.11 GiB | 6.546 | +-0% |
| nemotron-3.5-asr-streaming-0.6b | 0.58 GiB | 7.791 | +19% |
| parakeet-unified-en-0.6b | 0.56 GiB | 7.792 | +19% |
| Qwen3.5-4B | 3.28 GiB | 6.053 | +-8% |
| Qwythos-9B-Claude-Mythos-5-1M | 7.04 GiB | 6.426 | +-2% |
| Hy3 | 239.58 GiB | 6.888 | +5% |
| gemma-4-E4B-it | 5.79 GiB | 6.220 | +-5% |
| Qwen3-Coder-30B-A3B-Instruct | 23.37 GiB | 6.575 | +0% |
| gemma-4-31B-it | 23.47 GiB | 6.447 | +-2% |
| Qwen3-VL-30B-A3B-Instruct | 23.37 GiB | 6.461 | +-2% |
| FLUX.2-klein-9B | 7.33 GiB | 6.931 | +6% |
| cohere-transcribe-03-2026 | 1.85 GiB | 7.673 | +17% |
| LTX-2.3 | 16.55 GiB | 15.461 | +136% |
| Llama-3.2-1B-Instruct | 0.95 GiB | 6.615 | +1% |
| gemma-4-E2B-it | 4.19 GiB | 7.030 | +7% |
| gpt-oss-20b | 11.21 GiB | 4.478 | +-32% |
| Qwen3-8B | 6.26 GiB | 6.569 | +0% |
| Qwopus3.6-35B-A3B-v1 | 27.20 GiB | 6.500 | +-1% |
| Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP | 88.00 GiB | 27.210 | +315% |
| Qwen3.5-0.8B | 0.59 GiB | 5.768 | +-12% |
| Qwen3-VL-8B-Instruct-abliterated-v1 | 6.26 GiB | 6.137 | +-6% |
●From the filewhat these mean