The GPU memory each model needs in its common quantisation with an 8k context, and the cheapest GPU that fits it right now.
1 GB here is 1024³ bytes, like GPU memory. Hugging Face lists decimal gigabytes, so its numbers are 7% larger.
| Model | Parameters | Weights | VRAM | Cheapest GPU | Price |
|---|---|---|---|---|---|
| Qwen3.8 27B | 27.8B | Q4_K_M | 17.2 GB | 2× RTX 3060 12 GB | $0.23/h |
| Qwen3.5 9B | 9.7B | Q4_K_M | 6.8 GB | RTX 3060 12 GB | $0.115/h |
| Qwen3.5 4B | 4.7B | Q4_K_M | 4.1 GB | RTX A2000 6 GB | $0.09/h |
| Qwen3 8B | 8.2B | Q4_K_M | 6.8 GB | RTX 3060 12 GB | $0.115/h |
| Qwen3 14B | 14.8B | Q4_K_M | 10.7 GB | RTX 3060 12 GB | $0.115/h |
| Qwen3 32B | 32.8B | Q4_K_M | 21.4 GB | 2× RTX 3060 12 GB | $0.23/h |
| Qwen3 30B-A3B | 30.5B3.3B active | Q4_K_M | 19.1 GB | 2× RTX 3060 12 GB | $0.23/h |
| Qwen3 Coder 30B-A3B | 30.5B3.3B active | Q4_K_M | 19.1 GB | 2× RTX 3060 12 GB | $0.23/h |
| Qwen3 Next 80B-A3B | 81.3B3B active | Q4_K_M | 46.5 GB | RTX A6000 48 GB | $0.687/h |
| Qwen3 235B-A22B | 235B22B active | Q4_K_M | 134.9 GB | 2× A800 PCIe 80 GB | $2.08/h |
| Qwen3 Coder 480B-A35B | 480B35B active | Q4_K_M | 273.1 GB | 4× RTX PRO 5000 72 GB | $4.50/h |
| Gemma 4 12B | 12B | Q4_K_M | 8.3 GB | RTX 3060 12 GB | $0.115/h |
| Gemma 4 26B-A4B | 25.8B4B active | Q4_K_M | 17.4 GB | 2× RTX 3060 12 GB | $0.23/h |
| Gemma 4 31B | 31.3B | Q4_K_M | 19.7 GB | 2× RTX 3060 12 GB | $0.23/h |
| Gemma 3 27B | 27.4B | Q4_K_M | 17.7 GB | 2× RTX 3060 12 GB | $0.23/h |
| GPT-OSS 20B | 20.9B3.6B active | MXFP4 | 12.6 GB | RTX A4000 16 GB | $0.204/h |
| GPT-OSS 120B | 117B5.1B active | MXFP4 | 60.5 GB | 2× Tesla V100 32 GB | $0.819/h |
| DeepSeek R1 671B | 685B37B active | Q4_K_M | 378.7 GB | 8× RTX PRO 5000 72 GB | $9.00/h |
| DeepSeek V3.2 | 685B37B active | Q4_K_M | 379.1 GB | 8× RTX PRO 5000 72 GB | $9.00/h |
| DeepSeek V4 Flash | 291B | UD-Q4_K_XL | 145.5 GB | 2× A800 PCIe 80 GB | $2.08/h |
| DeepSeek R1 Distill Qwen 32B | 32.8B | Q4_K_M | 21.5 GB | 2× RTX 3060 12 GB | $0.23/h |
| DeepSeek R1 0528 Qwen3 8B | 8.2B | Q4_K_M | 6.8 GB | RTX 3060 12 GB | $0.115/h |
| Llama 3.1 8B | 8B | Q4_K_M | 6.6 GB | RTX 3060 12 GB | $0.115/h |
| Llama 3.3 70B | 70.6B | Q4_K_M | 43.1 GB | 4× RTX 3060 12 GB | $0.459/h |
| Mistral Small 3.2 24B | 24B | Q4_K_M | 15.6 GB | RTX A4000 16 GB | $0.204/h |
| Devstral Small 2 24B | 24B | Q4_K_M | 15.6 GB | RTX A4000 16 GB | $0.204/h |
| Ministral 3 14B | 14B | Q4_K_M | 9.9 GB | RTX 3060 12 GB | $0.115/h |
| Mistral Nemo 12B | 12.3B | Q4_K_M | 9.2 GB | RTX 3060 12 GB | $0.115/h |
| GLM-4.7 Flash | 31.2B3B active | Q4_K_M | 18.5 GB | 2× RTX 3060 12 GB | $0.23/h |
| GLM-4.5 Air | 110B12B active | Q4_K_M | 70.4 GB | A800 PCIe 80 GB | $1.04/h |
| GLM-5.3 Flash | 321B | UD-Q4_K_XL | 187.3 GB | 8× Tesla V100 32 GB | $3.28/h |
| Kimi K2 1T | 1,026B32B active | Q4_K_M | 580.2 GB | 8× RTX PRO 6000 96 GB | $18.25/h |
| Hermes 4.3 36B | 36.2B | Q4_K_M | 23.3 GB | RTX 3090 24 GB | $0.32/h |
| Phi-4 14B | 14.7B | Q4_K_M | 10.8 GB | RTX 3060 12 GB | $0.115/h |
| Hunyuan Hy3 | 299B | FP8 | 282.8 GB | 4× A100 SXM4 80 GB | $6.04/h |
| Model | Parameters | Weights | VRAM | Cheapest GPU | Price |
|---|---|---|---|---|---|
| FLUX.1 [dev] / [schnell] / Kontext | 11.9B | FP8 | 13.3 GB | RTX A4000 16 GB | $0.204/h |
| FLUX.2 [dev] | 32.2B | FP8 | 36.1 GB | RTX A6000 48 GB | $0.687/h |
| FLUX.2 [klein] 9B | 9.1B | FP8 | 10.6 GB | RTX 3060 12 GB | $0.115/h |
| FLUX.2 [klein] 4B | 3.9B | FP8 | 5.3 GB | RTX A2000 6 GB | $0.09/h |
| Qwen-Image / Qwen-Image-Edit |
| Model | Parameters | Weights | VRAM | Cheapest GPU | Price |
|---|---|---|---|---|---|
| Wan 2.2 A14B (T2V / I2V) | 14.3B | FP8 | 21.0 GB | RTX 3090 24 GB | $0.32/h |
| Wan 2.2 TI2V 5B | 5B | FP8 | 11.6 GB | RTX 3060 12 GB | $0.115/h |
| Wan 2.1 14B | 14.3B | FP8 | 21.0 GB | RTX 3090 24 GB | $0.32/h |
| Wan 2.1 1.3B | 1.4B | BF16 | 11.1 GB | RTX 3060 12 GB | $0.115/h |
| HunyuanVideo 1.5 |
| 20.4B |
| FP8 |
| 21.6 GB |
| RTX 3090 24 GB |
| $0.32/h |
| Stable Diffusion 3.5 Large | 8.2B | FP8 | 9.6 GB | RTX 3060 12 GB | $0.115/h |
| Stable Diffusion XL | 2.6B | BF16 | 6.8 GB | RTX 3060 12 GB | $0.115/h |
| Stable Diffusion 1.5 | 0.9B | BF16 | 2.9 GB | RTX A2000 6 GB | $0.09/h |
| 8.3B |
| FP8 |
| 18.5 GB |
| RTX 4000 Ada 20 GB |
| $0.313/h |
| LTX-2.5 22B | 21B | FP8 | 27.0 GB | Tesla V100 32 GB | $0.41/h |