Quantization format
IQ3_M
IQ3_M has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 906 files measured
Nominal bpw
—
from the block layout
Measured average
3.871
906 files
Range
2.453–13.190
varies by architecture
File sizes
0.03 GiB+
up to 454.54 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-35B-A3B | 16.57 GiB | 3.959 | — |
| gemma-4-26B-A4B-it | 12.37 GiB | 4.003 | — |
| gemma-4-12B-it | 5.34 GiB | 3.836 | — |
| Qwen3.5-4B | 2.31 GiB | 4.263 | — |
| Hy3 | 133.80 GiB | 3.847 | — |
| gemma-4-E4B-it | 4.39 GiB | 4.717 | — |
| gemma-4-31B-it | 14.09 GiB | 3.870 | — |
| Llama-3.2-1B-Instruct | 0.61 GiB | 4.255 | — |
| gemma-4-E2B-it | 2.92 GiB | 4.895 | — |
| Qwen3-8B | 3.63 GiB | 3.806 | — |
| Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP | 26.65 GiB | 8.239 | — |
| Qwen3.5-0.8B | 0.47 GiB | 4.649 | — |
| Llama-3.1-8B-Instruct | 3.52 GiB | 3.771 | — |
| Ornith-1.0-35B | 15.74 GiB | 3.901 | — |
| Qwopus3.6-27B-Coder | 12.14 GiB | 3.753 | — |
| Llama-3.2-3B-Instruct | 1.49 GiB | 3.983 | — |
| ThinkingCap-Qwen3.6-27B | 12.95 GiB | 4.066 | — |
| Qwythos-9B-v2 | 4.53 GiB | 4.027 | — |
| Qwen3.5-122B-A10B | 57.41 GiB | 3.943 | — |
| Qwen3-Coder-Next | 34.13 GiB | 3.679 | — |
| Qwen2.5-32B-Instruct | 13.79 GiB | 3.616 | — |
| Qwen3.5-35B-A3B | 16.57 GiB | 3.959 | — |
| Qwen2.5-Coder-7B-Instruct | 3.33 GiB | 3.754 | — |
| Qwen2.5-7B-Instruct | 3.33 GiB | 3.754 | — |
| Qwen3-0.6B | 0.38 GiB | 4.288 | — |
●From the filewhat these mean