Quantization publisher
city96
city96 publishes 241 quantizations across 19 models in our index, averaging 6.735 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 25 of the pairs we can compare — the same label does not mean the same file.
From the file· summed file bytes
Repositories
19
Quantizations
241
Models covered
19
Avg effective bpw
6.735
across their files
Same model, same quant label, different bytes
largest disagreements first
| Model | Quant | city96 | vs | Theirs | Difference |
|---|---|---|---|---|---|
| FLUX.2-dev | Q8_0 | 32.60 GiB | gguf-org | 51.67 GiB | -36.9% |
| FLUX.2-dev | Q4_0 | 17.97 GiB | gguf-org | 29.28 GiB | -38.6% |
| HunyuanVideo | Q8_0 | 13.01 GiB | calcuis | 20.97 GiB | -37.9% |
| FLUX.2-dev | Q2_K | 11.98 GiB | gguf-org | 18.79 GiB | -36.3% |
| stable-diffusion-3.5-large | Q5_1 | 5.84 GiB | second-state | 9.57 GiB | -39.0% |
| stable-diffusion-3.5-large | Q5_0 | 5.38 GiB | second-state | 8.82 GiB | -39.1% |
| stable-diffusion-3.5-large | Q4_1 | 4.91 GiB | second-state | 8.07 GiB | -39.2% |
| stable-diffusion-3.5-large | Q4_0 | 4.44 GiB | second-state | 7.32 GiB | -39.3% |
| t5-v1_1-xxl | F16 | 8.87 GiB | chatpig | 10.40 GiB | -14.7% |
| FLUX.2-dev | Q5_K_S | 21.63 GiB | unsloth | 22.12 GiB | -2.2% |
| FLUX.2-dev | Q6_K | 25.51 GiB | gguf-org | 25.98 GiB | -1.8% |
| FLUX.2-dev | Q5_1 | 23.46 GiB | gguf-org | 23.92 GiB | -1.9% |
| FLUX.2-dev | Q4_1 | 19.80 GiB | gguf-org | 20.27 GiB | -2.3% |
| FLUX.2-dev | Q5_K_S | 21.63 GiB | gguf-org | 22.10 GiB | -2.1% |
| FLUX.2-dev | Q4_K_S | 17.97 GiB | gguf-org | 18.44 GiB | -2.5% |
| FLUX.2-dev | Q5_0 | 21.63 GiB | gguf-org | 22.10 GiB | -2.1% |
| FLUX.2-dev | Q4_K_S | 17.97 GiB | unsloth | 18.43 GiB | -2.5% |
| Qwen-Image | Q5_K_S | 13.15 GiB | unsloth | 13.32 GiB | -1.3% |
| Qwen-Image | Q2_K | 6.58 GiB | unsloth | 6.72 GiB | -2.2% |
| FLUX.2-dev | Q3_K_S | 14.70 GiB | gguf-org | 14.55 GiB | +1.0% |
| FLUX.2-dev | Q5_K_M | 22.41 GiB | unsloth | 22.28 GiB | +0.5% |
| FLUX.2-dev | Q3_K_M | 14.86 GiB | unsloth | 14.74 GiB | +0.8% |
| FLUX.2-dev | Q3_K_S | 14.70 GiB | unsloth | 14.57 GiB | +0.8% |
| Qwen-Image | Q4_K_S | 11.31 GiB | unsloth | 11.43 GiB | -1.0% |
| FLUX.2-dev | Q4_K_M | 18.70 GiB | unsloth | 18.59 GiB | +0.6% |
A quantization label describes a target, not a recipe. Publishers make different choices about which tensors to keep at higher precision, and some apply an importance matrix while others don't — so two files both honestly labelled the same thing can differ measurably in size and in quality.
Models they publish
| Model | Quantizations | Smallest |
|---|---|---|
| FLUX.1-dev | 11 | 3.76 GiB |
| FLUX.1-schnell | 11 | 3.73 GiB |
| FLUX.2-dev | 14 | 11.98 GiB |
| umt5-xxl | 10 | 2.66 GiB |
| Qwen-Image | 14 | 6.58 GiB |
| t5-v1_1-xxl | 11 | 1.96 GiB |
| Wan2.1-I2V-14B-480P | 14 | 7.38 GiB |
| stable-diffusion-3.5-large | 6 | 4.44 GiB |
| LTX-Video | 15 | 0.92 GiB |
| Wan2.1-T2V-14B | 14 | 6.51 GiB |
| stable-diffusion-3.5-medium | 13 | 1.35 GiB |
| stable-diffusion-3-medium | 11 | 1.20 GiB |
| Wan2.1-I2V-14B-720P | 14 | 7.38 GiB |
| HiDream-I1-Full | 14 | 6.11 GiB |
| Wan2.1-FLF2V-14B-720P | 14 | 7.38 GiB |
| HunyuanVideo | 13 | 5.67 GiB |
| llava-llama-3-8b-v1_1-transformers | 15 | 3.41 GiB |
| HunyuanVideo-I2V | 13 | 5.67 GiB |
| AuraFlow-v0.3 | 14 | 2.23 GiB |