runlocal sizes models from measured GGUF bytes where we have them, not a flat estimate. This tracks how far the naïve bits-per-weight prediction drifts from the real on-disk size — across 96 measured quants, the flat estimate is off by 2.0% on average.
| model | quant | predicted | measured | error |
|---|---|---|---|---|
| DeepSeek-R1 (distill) 70B (Llama) | Q2_K | 27.5 GB | 24.6 GB | +12.0% |
| Llama 3.1 70B | Q2_K | 27.5 GB | 24.6 GB | +12.0% |
| Llama 3.3 70B | Q2_K | 27.5 GB | 24.6 GB | +12.0% |
| Nemotron-Mini 4B 4B | Q4_K_M | 2.3 GB | 2.5 GB | -10.1% |
| Llama 3.2 1B | Q4_K_M | 715 MB | 770 MB | -7.2% |
| Falcon3 7B 7B | Q4_K_M | 4.0 GB | 4.3 GB | -7.1% |
| Falcon3 7B 7B | Q5_K_M | 4.6 GB | 5.0 GB | -6.4% |
| Falcon3 7B 7B | Q6_K | 5.3 GB | 5.7 GB | -6.2% |
| Falcon3 7B 7B | Q8_0 | 6.9 GB | 7.4 GB | -6.2% |
| OLMo 2 7B 7B | Q4_K_M | 4.0 GB | 4.2 GB | -5.1% |
| Gemma 4 12B | Q4_K_M | 6.8 GB | 7.1 GB | -5.1% |
| SmolLM3 3B 3B | Q4_K_M | 1.7 GB | 1.8 GB | -5.0% |
| Nemotron-Mini 4B 4B | Q6_K | 3.1 GB | 3.2 GB | -4.8% |
| Nemotron-Mini 4B 4B | Q8_0 | 4.0 GB | 4.2 GB | -4.7% |
| OLMo 2 7B 7B | Q6_K | 5.3 GB | 5.6 GB | -4.2% |
| OLMo 2 7B 7B | Q8_0 | 6.9 GB | 7.2 GB | -4.2% |
| Gemma 4 12B | Q6_K | 9.2 GB | 9.5 GB | -3.9% |
| SmolLM3 3B 3B | Q5_K_M | 2.0 GB | 2.1 GB | -3.6% |
| DeepSeek-R1 (distill) 1.5B | Q4_K_M | 1.0 GB | 1.0 GB | -3.6% |
| Llama 3.2 3B | Q4_K_M | 1.8 GB | 1.9 GB | -3.5% |
| Gemma 4 12B | Q8_0 | 11.9 GB | 12.2 GB | -2.9% |
| Gemma 4 12B | Q5_K_M | 7.9 GB | 8.2 GB | -2.7% |
| Phi-4 14B | Q3_K_M | 6.7 GB | 6.9 GB | -2.7% |
| SmolLM3 3B 3B | Q8_0 | 3.0 GB | 3.1 GB | -2.7% |
| Yi-Coder 9B 9B | Q4_K_M | 5.1 GB | 5.0 GB | +2.4% |
| QwQ 32B 32B | Q8_0 | 31.7 GB | 32.4 GB | -2.4% |
| QwQ 32B 32B | Q4_K_M | 18.1 GB | 18.5 GB | -2.3% |
| Granite 3.3 8B 8B | Q6_K | 6.1 GB | 6.2 GB | -2.2% |
| Granite 3.1 8B 8B | Q6_K | 6.1 GB | 6.2 GB | -2.2% |
| QwQ 32B 32B | Q5_K_M | 21.2 GB | 21.7 GB | -2.2% |
| Granite 3.3 8B 8B | Q8_0 | 7.9 GB | 8.1 GB | -2.1% |
| Granite 3.1 8B 8B | Q8_0 | 7.9 GB | 8.1 GB | -2.1% |
| Yi-Coder 9B 9B | Q8_0 | 8.9 GB | 8.7 GB | +1.9% |
| Granite 3.3 8B 8B | Q4_K_M | 4.5 GB | 4.6 GB | -1.9% |
| Granite 3.1 8B 8B | Q4_K_M | 4.5 GB | 4.6 GB | -1.9% |
| Yi-Coder 9B 9B | Q6_K | 6.9 GB | 6.7 GB | +1.9% |
| Granite 3.3 8B 8B | Q5_K_M | 5.3 GB | 5.4 GB | -1.9% |
| Granite 3.1 8B 8B | Q5_K_M | 5.3 GB | 5.4 GB | -1.9% |
| Phi-4 14B | Q4_K_M | 8.3 GB | 8.4 GB | -1.8% |
| Phi-4 14B | Q5_K_M | 9.7 GB | 9.9 GB | -1.7% |
| DeepSeek-R1 (distill) 14B | Q3_K_M | 6.7 GB | 6.8 GB | -1.6% |
| Llama 3.2 3B | Q5_K_M | 2.1 GB | 2.2 GB | -1.6% |
| Qwen2.5-Coder 7B | Q4_K_M | 4.3 GB | 4.4 GB | -1.4% |
| DeepSeek-R1 (distill) 7B | Q4_K_M | 4.3 GB | 4.4 GB | -1.4% |
| Llama 3.1 8B | Q4_K_M | 4.5 GB | 4.6 GB | -1.1% |
| DeepSeek-R1 (distill) 8B (Llama) | Q4_K_M | 4.5 GB | 4.6 GB | -1.1% |
| Qwen3 Coder 30B-A3B 30B-A3B (MoE) | Q4_K_M | 17.2 GB | 17.4 GB | -0.7% |
| Llama 3.1 70B | Q3_K_M | 32.1 GB | 31.9 GB | +0.6% |
| Llama 3.3 70B | Q3_K_M | 32.1 GB | 31.9 GB | +0.6% |
| Llama 3.2 1B | Q8_0 | 1.2 GB | 1.2 GB | -0.6% |
| DeepSeek-R1 (distill) 70B (Llama) | Q4_K_M | 39.8 GB | 39.6 GB | +0.6% |
| Llama 3.1 70B | Q4_K_M | 39.8 GB | 39.6 GB | +0.6% |
| Llama 3.3 70B | Q4_K_M | 39.8 GB | 39.6 GB | +0.6% |
| Qwen2.5-Coder 7B | Q5_K_M | 5.0 GB | 5.1 GB | -0.5% |
| DeepSeek-R1 (distill) 7B | Q5_K_M | 5.0 GB | 5.1 GB | -0.5% |
| DeepSeek-R1 (distill) 32B | Q3_K_M | 14.9 GB | 14.8 GB | +0.5% |
| Qwen2.5-Coder 32B | Q3_K_M | 14.9 GB | 14.8 GB | +0.5% |
| Mistral 7B 7B | Q4_K_M | 4.1 GB | 4.1 GB | +0.5% |
| DeepSeek-R1 (distill) 70B (Llama) | Q5_K_M | 46.7 GB | 46.5 GB | +0.5% |
| Llama 3.1 70B | Q5_K_M | 46.7 GB | 46.5 GB | +0.5% |
| Llama 3.3 70B | Q5_K_M | 46.7 GB | 46.5 GB | +0.5% |
| Mistral Small 3 24B | Q3_K_M | 10.7 GB | 10.7 GB | +0.4% |
| DeepSeek-R1 (distill) 14B | Q4_K_M | 8.3 GB | 8.4 GB | -0.4% |
| Llama 3.1 8B | Q5_K_M | 5.3 GB | 5.3 GB | -0.4% |
| DeepSeek-R1 (distill) 8B (Llama) | Q5_K_M | 5.3 GB | 5.3 GB | -0.4% |
| Mistral 7B 7B | Q5_K_M | 4.8 GB | 4.8 GB | +0.4% |
| DeepSeek-R1 (distill) 1.5B | Q8_0 | 1.8 GB | 1.8 GB | -0.3% |
| Llama 3.2 1B | FP16 | 2.3 GB | 2.3 GB | -0.3% |
| Mistral Small 3 24B | Q4_K_M | 13.3 GB | 13.3 GB | -0.3% |
| Qwen3 Coder 30B-A3B 30B-A3B (MoE) | Q6_K | 23.3 GB | 23.4 GB | -0.3% |
| Llama 3.2 3B | Q8_0 | 3.2 GB | 3.2 GB | -0.2% |
| DeepSeek-R1 (distill) 32B | Q5_K_M | 21.7 GB | 21.7 GB | +0.2% |
| Qwen2.5-Coder 32B | Q5_K_M | 21.7 GB | 21.7 GB | +0.2% |
| DeepSeek-R1 (distill) 1.5B | FP16 | 3.3 GB | 3.3 GB | -0.2% |
| Llama 3.1 8B | Q6_K | 6.1 GB | 6.1 GB | -0.2% |
| Qwen2.5-Coder 7B | Q6_K | 5.8 GB | 5.8 GB | -0.2% |
| DeepSeek-R1 (distill) 7B | Q6_K | 5.8 GB | 5.8 GB | -0.2% |
| Qwen3 Coder 30B-A3B 30B-A3B (MoE) | Q8_0 | 30.2 GB | 30.3 GB | -0.1% |
| Llama 3.1 8B | Q8_0 | 7.9 GB | 8.0 GB | -0.1% |
| DeepSeek-R1 (distill) 8B (Llama) | Q8_0 | 7.9 GB | 8.0 GB | -0.1% |
| Qwen2.5-Coder 7B | Q8_0 | 7.5 GB | 7.5 GB | -0.1% |
| DeepSeek-R1 (distill) 7B | Q8_0 | 7.5 GB | 7.5 GB | -0.1% |
| Mistral 7B 7B | Q6_K | 5.5 GB | 5.5 GB | -0.1% |
| DeepSeek-R1 (distill) 32B | Q4_K_M | 18.5 GB | 18.5 GB | +0.1% |
| Qwen2.5-Coder 32B | Q4_K_M | 18.5 GB | 18.5 GB | +0.1% |
| DeepSeek-R1 (distill) 14B | Q8_0 | 14.6 GB | 14.6 GB | -0.1% |
| Mistral Small 3 24B | Q8_0 | 23.3 GB | 23.3 GB | -0.0% |
| DeepSeek-R1 (distill) 14B | Q5_K_M | 9.8 GB | 9.8 GB | -0.0% |
| Phi-4 14B | Q8_0 | 14.5 GB | 14.5 GB | -0.0% |
| Qwen2.5-Coder 32B | Q8_0 | 32.4 GB | 32.4 GB | -0.0% |
| DeepSeek-R1 (distill) 32B | Q8_0 | 32.4 GB | 32.4 GB | -0.0% |
| Mistral 7B 7B | Q8_0 | 7.2 GB | 7.2 GB | -0.0% |
| Llama 3.3 70B | Q8_0 | 69.8 GB | 69.8 GB | -0.0% |
| Llama 3.1 70B | Q8_0 | 69.8 GB | 69.8 GB | -0.0% |
| DeepSeek-R1 (distill) 70B (Llama) | Q8_0 | 69.8 GB | 69.8 GB | -0.0% |
| Mistral Small 3 24B | Q5_K_M | 15.6 GB | 15.6 GB | +0.0% |
Positive error = the flat estimate runs heavy vs the real file. Browse the full catalog →