VRAM Requirements by Model
Memory needed by 58 language models and 15 image and video models. LLM figures are at Q4_K_M (or the official native format) with an 8K context; image figures use FP8 weights with model swapping.
Language models
| Model | Parameters | Q4_K_M, 8K | Q8_0, 8K | Smallest tier |
|---|---|---|---|---|
| Gemma 3 1B | 1B | 1.6 GB | 2.0 GB | 4 GB |
| Llama 3.2 1B | 1.24B | 1.7 GB | 2.3 GB | 4 GB |
| Llama 3.2 3B | 3.21B | 3.5 GB | 4.8 GB | 4 GB |
| Ministral 3 3B | 3.43B | 3.5 GB | 5.0 GB | 4 GB |
| Phi-4 Mini | 3.84B | 4.1 GB | 5.7 GB | 6 GB |
| Qwen 3 4B | 4.02B | 4.2 GB | 5.9 GB | 6 GB |
| Qwen 3.5 4B | 4.2B | 3.7 GB | 5.5 GB | 4 GB |
| Gemma 3 4B | 4.3B | 3.7 GB | 5.6 GB | 4 GB |
| Gemma 4 E2B | 5.1B | 4.0 GB | 6.1 GB | 6 GB |
| Mistral 7B v0.3 | 7.25B | 5.7 GB | 8.8 GB | 8 GB |
| Qwen 2.5 7B | 7.62B | 5.6 GB | 8.8 GB | 6 GB |
| DeepSeek R1 Distill Qwen 7B | 7.62B | 5.6 GB | 8.8 GB | 6 GB |
| Gemma 4 E4B | 8B | 5.7 GB | 9.1 GB | 8 GB |
| Llama 3.1 8B | 8.03B | 6.3 GB | 9.8 GB | 8 GB |
| DeepSeek R1 Distill Llama 8B | 8.03B | 6.3 GB | 9.8 GB | 8 GB |
| Qwen 3 8B | 8.19B | 6.6 GB | 10.1 GB | 8 GB |
| Ministral 3 8B | 8.49B | 6.7 GB | 10.3 GB | 8 GB |
| Qwen 3.5 9B | 8.95B | 6.4 GB | 10.2 GB | 8 GB |
| Mistral Nemo 12B | 12.2B | 9.0 GB | 14.2 GB | 10 GB |
| Gemma 3 12B | 12.2B | 8.8 GB | 13.9 GB | 10 GB |
| Ministral 3 14B | 13.5B | 9.7 GB | 15.4 GB | 12 GB |
| Qwen 2.5 14B | 14.7B | 10.7 GB | 16.9 GB | 12 GB |
| Phi-4 14B | 14.7B | 10.6 GB | 16.9 GB | 12 GB |
| DeepSeek R1 Distill Qwen 14B | 14.7B | 10.7 GB | 16.9 GB | 12 GB |
| Qwen 3 14B | 14.8B | 10.5 GB | 16.8 GB | 12 GB |
| gpt-oss-20b | 20.9B (3.6B active) | 13.9 GB | 21.8 GB | 16 GB |
| Mistral Small 3.1 24B | 24B | 15.6 GB | 25.8 GB | 20 GB |
| Gemma 4 26B-A4B | 25.2B (3.8B active) | 15.6 GB | 26.3 GB | 20 GB |
| Qwen 3.5 27B | 26.9B | 16.9 GB | 28.3 GB | 20 GB |
| Qwen 3.6 27B | 26.9B | 16.9 GB | 28.3 GB | 20 GB |
| Qwen 3.8 27B | 26.9B | 16.9 GB | 28.3 GB | 20 GB |
| Gemma 3 27B | 27.4B | 17.6 GB | 29.2 GB | 20 GB |
| Qwen 3 30B-A3B | 30.5B (3.3B active) | 18.8 GB | 31.8 GB | 20 GB |
| Qwen3-Coder 30B-A3B | 30.5B (3.3B active) | 18.8 GB | 31.8 GB | 20 GB |
| Gemma 4 31B | 30.7B | 19.8 GB | 32.9 GB | 24 GB |
| Qwen 2.5 32B | 32.8B | 21.4 GB | 35.3 GB | 24 GB |
| Qwen 2.5 Coder 32B | 32.8B | 21.4 GB | 35.3 GB | 24 GB |
| Qwen 3 32B | 32.8B | 21.4 GB | 35.3 GB | 24 GB |
| DeepSeek R1 Distill Qwen 32B | 32.8B | 21.4 GB | 35.3 GB | 24 GB |
| Qwen 3.5 35B-A3B | 35B (3B active) | 21.0 GB | 35.9 GB | 24 GB |
| Qwen 3.6 35B-A3B | 35B (3B active) | 21.0 GB | 35.9 GB | 24 GB |
| Mixtral 8x7B | 46.7B (12.9B active) | 28.0 GB | 47.8 GB | 32 GB |
| Llama 3.3 70B | 70.6B | 43.2 GB | 73.2 GB | 48 GB |
| DeepSeek R1 Distill Llama 70B | 70.6B | 43.2 GB | 73.2 GB | 48 GB |
| Qwen 2.5 72B | 72.7B | 44.5 GB | 75.4 GB | 48 GB |
| Qwen3-Next 80B-A3B | 80B (3B active) | 46.3 GB | 80.2 GB | 64 GB |
| GLM-4.5-Air | 106B (12B active) | 62.1 GB | 107 GB | 80 GB |
| Llama 4 Scout | 109B (17B active) | 64.0 GB | 110 GB | 80 GB |
| gpt-oss-120b | 116.8B (5.1B active) | 62.0 GB | 117 GB | 80 GB |
| Mistral Small 4 119B | 119B (6.5B active) | 112 GB | 119 GB | 128 GB |
| Qwen 3.5 122B-A10B | 122B (10B active) | 70.2 GB | 122 GB | 80 GB |
| MiniMax-M2 | 230B (10B active) | 218 GB | 230 GB | Multi-GPU |
| Qwen 3 235B-A22B | 235.1B (22B active) | 135 GB | 235 GB | Multi-GPU |
| DeepSeek V4 Flash | 284B (13B active) | 144 GB | 282 GB | Multi-GPU |
| Qwen 3.5 397B-A17B | 397B (17B active) | 226 GB | 394 GB | Multi-GPU |
| Llama 3.1 405B | 405.9B | 234 GB | 407 GB | Multi-GPU |
| DeepSeek R1 671B | 671B (37B active) | 642 GB | 665 GB | Multi-GPU |
| Kimi K2.6 | 1000B (32B active) | 556 GB | 991 GB | Multi-GPU |
Image and video models
| Model | Type | Diffusion params | FP8 peak |
|---|---|---|---|
| Stable Diffusion 1.5 | Image | 0.86B | 1.8 GB |
| SDXL 1.0 | Image | 2.57B | 4.1 GB |
| Stable Diffusion 3.5 Medium | Image | 2.5B | 6.4 GB |
| Stable Diffusion 3.5 Large | Image | 8.1B | 8.6 GB |
| FLUX.1 [dev] | Image | 11.9B | 12.1 GB |
| FLUX.1 [schnell] | Image | 11.9B | 12.1 GB |
| Qwen-Image | Image | 20.4B | 20.4 GB |
| Wan 2.2 A14B (T2V / I2V) | Video | 14B | 18.8 GB |
| Wan 2.2 TI2V 5B | Video | 5B | 8.0 GB |
| HunyuanVideo | Video | 13B | 18.3 GB |
| FLUX.1 Kontext [dev] | Image | 11.9B | 12.3 GB |
| FLUX.2 [dev] | Image | 32B | 31.0 GB |
| Z-Image Turbo | Image | 6.15B | 6.8 GB |
| Qwen-Image-Edit 2509 | Image | 20.4B | 20.5 GB |
| LTX-2 | Video | 19B | 19.2 GB |
Missing a model? It usually appears here within days of release. Request one.