LLM VRAM requirements by model

Pick a model to see how much memory each quant needs and which GPUs it fits. Each card shows the default quant and its estimated total at 8K context. Search by name, or jump to a family.

Popular: Qwen3-8B · Qwen3-VL-8B-Instruct ·Qwen3.8-27B · Qwen3.8-Flash-Next ·Qwen3-8B on an RTX 4070

Search models

86 models

Current models

DeepSeek-V4.1-Flash763B · MoE · DeepSeekFP8 safetensors · 475.9 GiB–548.0 GiBdeepseek-ai/DeepSeek-V4.1-Flash GLM-5.3-Flash321B · MoE · GLMGGUF Q8_0 (Unsloth) · 318.0 GiB–373.6 GiBzai-org/GLM-5.3-Flash Qwen3.8-27B28B · Dense · QwenGGUF Q4_K_M (mradermacher) · 16.4 GiB–19.7 GiBQwen/Qwen3.8-27B Qwen3.8-Flash-Next180B · MoE · QwenGGUF Q4_K_M (bartowski) · 111.8 GiB–131.5 GiBQwen/Qwen3.8-Flash-Next

All models by family

Qwen 39

Qwen3-VL-8B-Instruct8.8B · Dense · 2025-10-11GGUF Q4_K_M (Qwen) · 5.9 GiB–6.5 GiBQwen/Qwen3-VL-8B-Instruct Qwen3-VL-4B-Instruct4.4B · Dense · 2025-10-11GGUF Q4_K_M (Qwen) · 3.6 GiB–4.1 GiBQwen/Qwen3-VL-4B-Instruct Qwen3-VL-30B-A3B-Instruct31B · 3.0B active · MoE · 2025-09-30GGUF Q4_K_M (Qwen) · 18.2 GiB–18.9 GiBQwen/Qwen3-VL-30B-A3B-Instruct Qwen3-Next-80B-A3B-Instruct81B · 3.0B active · MoE · 2025-09-09GGUF Q4_K_M (Qwen) · 46.0 GiB–47.3 GiBQwen/Qwen3-Next-80B-A3B-Instruct Qwen3-4B-Instruct-25074.0B · Dense · 2025-08-05GGUF Q4_K_M (Unsloth) · 3.6 GiB–4.1 GiBQwen/Qwen3-4B-Instruct-2507 Qwen3-30B-A3B-Instruct-250731B · 3.3B active · MoE · 2025-07-28GGUF Q4_K_M (Unsloth) · 18.2 GiB–18.9 GiBQwen/Qwen3-30B-A3B-Instruct-2507 Qwen3-235B-A22B-Instruct-2507235B · 22B active · MoE · 2025-07-21GGUF Q4_K_M (Unsloth) · 134.0 GiB–137.1 GiBQwen/Qwen3-235B-A22B-Instruct-2507 Qwen3-8B-Base8.2B · Dense · 2025-04-28GGUF Q4_K_M (mradermacher) · 5.9 GiB–6.5 GiBQwen/Qwen3-8B-Base Qwen3-1.7B-Base1.7B · Dense · 2025-04-28GGUF Q4_K_M (mradermacher) · 2.0 GiB–2.5 GiBQwen/Qwen3-1.7B-Base Qwen3-235B-A22B235B · 22B active · MoE · 2025-04-27GGUF Q4_K_M (Qwen) · 134.0 GiB–137.1 GiBQwen/Qwen3-235B-A22B Qwen3-32B33B · Dense · 2025-04-27GGUF Q4_K_M · 20.5 GiB–21.4 GiBQwen/Qwen3-32B Qwen3-30B-A3B31B · 3.3B active · MoE · 2025-04-27GGUF Q4_K_M · 18.2 GiB–18.9 GiBQwen/Qwen3-30B-A3B Qwen3-14B15B · Dense · 2025-04-27GGUF Q4_K_M · 9.8 GiB–10.4 GiBQwen/Qwen3-14B Qwen3-8B8.2B · Dense · 2025-04-27GGUF Q4_K_M · 5.9 GiB–6.5 GiBQwen/Qwen3-8B Qwen3-4B4.0B · Dense · 2025-04-27GGUF Q4_K_M · 3.6 GiB–4.1 GiBQwen/Qwen3-4B Qwen3-1.7B2.0B · Dense · 2025-04-27GGUF Q4_K_M (Unsloth) · 2.0 GiB–2.5 GiBQwen/Qwen3-1.7B Qwen3-0.6B0.8B · Dense · 2025-04-27GGUF Q4_K_M (Unsloth) · 1.4 GiB–1.8 GiBQwen/Qwen3-0.6B OpenMath-Nemotron-32B33B · Dense · 2025-04-22GGUF Q4_K_M (mradermacher) · 20.6 GiB–21.5 GiBnvidia/OpenMath-Nemotron-32B OpenMath-Nemotron-14B15B · Dense · 2025-04-22GGUF Q4_K_M (mradermacher) · 10.0 GiB–10.6 GiBnvidia/OpenMath-Nemotron-14B OpenMath-Nemotron-7B7.6B · Dense · 2025-04-22GGUF Q4_K_M (mradermacher) · 4.9 GiB–5.4 GiBnvidia/OpenMath-Nemotron-7B Qwen2.5-VL-32B-Instruct33B · Dense · 2025-03-21GGUF Q4_K_M (Unsloth) · 20.6 GiB–21.5 GiBQwen/Qwen2.5-VL-32B-Instruct QwQ-32B33B · Dense · 2025-03-05GGUF Q4_K_M (Qwen) · 20.6 GiB–21.5 GiBQwen/QwQ-32B Qwen2.5-VL-7B-Instruct8.3B · Dense · 2025-01-26GGUF Q4_K_M (Unsloth) · 4.9 GiB–5.4 GiBQwen/Qwen2.5-VL-7B-Instruct Qwen2.5-VL-3B-Instruct3.8B · Dense · 2025-01-26GGUF Q4_K_M (Unsloth) · 2.2 GiB–2.6 GiBQwen/Qwen2.5-VL-3B-Instruct Qwen2.5-Coder-32B-Instruct33B · Dense · 2024-11-06GGUF Q4_K_M (Qwen) · 20.6 GiB–21.5 GiBQwen/Qwen2.5-Coder-32B-Instruct Qwen2.5-Coder-14B-Instruct15B · Dense · 2024-11-06GGUF Q4_K_M (Qwen) · 10.0 GiB–10.6 GiBQwen/Qwen2.5-Coder-14B-Instruct Qwen2.5-Coder-3B-Instruct3.1B · Dense · 2024-11-06GGUF Q4_K_M (Qwen) · 2.4 GiB–2.8 GiBQwen/Qwen2.5-Coder-3B-Instruct Qwen2.5-Math-7B-Instruct7.6B · Dense · 2024-09-19GGUF Q4_K_M (bartowski) · 4.9 GiB–6.2 GiBQwen/Qwen2.5-Math-7B-Instruct Qwen2.5-Coder-1.5B-Instruct1.5B · Dense · 2024-09-18GGUF Q4_K_M (Qwen) · 1.4 GiB–1.8 GiBQwen/Qwen2.5-Coder-1.5B-Instruct Qwen2.5-32B-Instruct33B · Dense · 2024-09-17GGUF Q4_K_M · 20.6 GiB–21.5 GiBQwen/Qwen2.5-32B-Instruct Qwen2.5-Coder-7B-Instruct7.6B · Dense · 2024-09-17GGUF Q4_K_M (Qwen) · 4.9 GiB–5.4 GiBQwen/Qwen2.5-Coder-7B-Instruct Qwen2.5-3B-Instruct3.1B · Dense · 2024-09-17GGUF Q4_K_M · 2.4 GiB–2.8 GiBQwen/Qwen2.5-3B-Instruct Qwen2.5-1.5B-Instruct1.5B · Dense · 2024-09-17GGUF Q4_K_M (Qwen) · 1.4 GiB–1.8 GiBQwen/Qwen2.5-1.5B-Instruct Qwen2.5-72B-Instruct73B · Dense · 2024-09-16GGUF Q4_K_M · 43.6 GiB–45.0 GiBQwen/Qwen2.5-72B-Instruct Qwen2.5-14B-Instruct15B · Dense · 2024-09-16GGUF Q4_K_M · 10.0 GiB–10.6 GiBQwen/Qwen2.5-14B-Instruct Qwen2.5-7B-Instruct7.6B · Dense · 2024-09-16GGUF Q4_K_M · 4.9 GiB–5.4 GiBQwen/Qwen2.5-7B-Instruct Qwen2.5-Coder-7B7.6B · Dense · 2024-09-16GGUF Q4_K_M (mradermacher) · 4.9 GiB–5.4 GiBQwen/Qwen2.5-Coder-7B Qwen2.5-0.5B-Instruct0.5B · Dense · 2024-09-16GGUF Q4_K_M (Qwen) · 0.7 GiB–1.1 GiBQwen/Qwen2.5-0.5B-Instruct Qwen2-7B-Instruct7.6B · Dense · 2024-06-04GGUF Q4_K_M (Qwen) · 4.9 GiB–5.4 GiBQwen/Qwen2-7B-Instruct

DeepSeek 10

DeepSeek-V4-Flash-0731304B · MoE · 2026-07-31FP8 safetensors · 156.2 GiB–180.2 GiBdeepseek-ai/DeepSeek-V4-Flash-0731 DeepSeek-V3.1685B · 37B active · MoE · 2025-08-21GGUF Q4_K_M (Unsloth) · 378.2 GiB–444.2 GiBdeepseek-ai/DeepSeek-V3.1 DeepSeek-R1-0528685B · 37B active · MoE · 2025-05-28GGUF Q4_K_M (Unsloth) · 377.8 GiB–443.7 GiBdeepseek-ai/DeepSeek-R1-0528 DeepSeek-V3-0324685B · 37B active · MoE · 2025-03-24GGUF Q4_K_M (Unsloth) · 377.3 GiB–443.1 GiBdeepseek-ai/DeepSeek-V3-0324 DeepSeek-R1684B · 37B active · MoE · 2025-01-20GGUF Q4_K_M (Unsloth) · 377.3 GiB–443.1 GiBdeepseek-ai/DeepSeek-R1 DeepSeek-R1-Distill-Qwen-32B33B · Dense · 2025-01-20GGUF Q4_K_M (Unsloth) · 20.6 GiB–21.5 GiBdeepseek-ai/DeepSeek-R1-Distill-Qwen-32B DeepSeek-R1-Distill-Llama-8B8.0B · Dense · 2025-01-20GGUF Q4_K_M (Unsloth) · 5.7 GiB–6.2 GiBdeepseek-ai/DeepSeek-R1-Distill-Llama-8B DeepSeek-R1-Distill-Qwen-7B7.6B · Dense · 2025-01-20GGUF Q4_K_M (Unsloth) · 4.9 GiB–5.4 GiBdeepseek-ai/DeepSeek-R1-Distill-Qwen-7B DeepSeek-V3685B · 37B active · MoE · 2024-12-25GGUF Q4_K_M (Unsloth) · 377.3 GiB–443.1 GiBdeepseek-ai/DeepSeek-V3 DeepSeek-V2-Lite-Chat16B · 2.4B active · MoE · 2024-05-15GGUF Q4_K_M (mradermacher) · 10.0 GiB–12.2 GiBdeepseek-ai/DeepSeek-V2-Lite-Chat

Llama 6

Llama 3.3 70B Instruct71B · Dense · 2024-12-06Official BF16 safetensors · 134.0 GiB–134.5 GiBmeta-llama/Llama-3.3-70B-Instruct SmolLM2-1.7B-Instruct1.7B · Dense · 2024-10-31GGUF Q4_K_M (Hugging Face) · 2.6 GiB–3.1 GiBHuggingFaceTB/SmolLM2-1.7B-Instruct Llama 3.2 3B Instruct3.2B · Dense · 2024-10-24Official BF16 safetensors · 6.9 GiB–7.5 GiBmeta-llama/Llama-3.2-3B-Instruct Llama 3.2 1B Instruct1.2B · Dense · 2024-10-24Official BF16 safetensors · 2.6 GiB–3.2 GiBmeta-llama/Llama-3.2-1B-Instruct Llama-3.1-Nemotron-70B-Instruct-HF71B · Dense · 2024-10-12GGUF Q4_K_M (bartowski) · 42.2 GiB–43.6 GiBnvidia/Llama-3.1-Nemotron-70B-Instruct-HF Llama 3.1 8B Instruct8.0B · Dense · 2024-07-23Official BF16 safetensors · 16.0 GiB–16.6 GiBmeta-llama/Llama-3.1-8B-Instruct

GLM 5

GLM-5.3753B · MoE · 2026-08-25GGUF Q8_0 (Unsloth) · 747.1 GiB–877.0 GiBzai-org/GLM-5.3 GLM-4.6357B · MoE · 2025-09-29GGUF Q4_K_M (Unsloth) · 203.8 GiB–208.4 GiBzai-org/GLM-4.6 GLM-4.5358B · MoE · 2025-07-20GGUF Q4_K_M (Unsloth) · 204.6 GiB–209.2 GiBzai-org/GLM-4.5 GLM-4.5-Air110B · MoE · 2025-07-20GGUF Q4_K_M (Unsloth) · 69.5 GiB–71.4 GiBzai-org/GLM-4.5-Air GLM-4-9B-04149.4B · Dense · 2025-04-07GGUF Q4_K_M (Unsloth) · 6.2 GiB–6.7 GiBzai-org/GLM-4-9B-0414

Google 4

Gemma 3 27B IT27B · Dense · 2025-03-10GGUF Q4_0 (Google QAT) · 17.2 GiB–18.0 GiBgoogle/gemma-3-27b-it Gemma 3 12B IT12B · Dense · 2025-03-10GGUF Q4_0 (Google QAT) · 8.5 GiB–9.0 GiBgoogle/gemma-3-12b-it Gemma 3 4B IT4.3B · Dense · 2025-03-10GGUF Q4_0 (Google QAT) · 3.3 GiB–3.8 GiBgoogle/gemma-3-4b-it Gemma 3 1B IT1.0B · Dense · 2025-03-10GGUF Q4_0 (Google QAT) · 1.1 GiB–1.5 GiBgoogle/gemma-3-1b-it

Mistral 4

Mistral-Small-24B-Instruct-250124B · Dense · 2025-01-28GGUF Q4_K_M (Unsloth) · 14.7 GiB–15.5 GiBmistralai/Mistral-Small-24B-Instruct-2501 Mistral-Nemo-Instruct-240712B · Dense · 2024-07-17GGUF Q4_K_M (Unsloth) · 8.3 GiB–8.9 GiBmistralai/Mistral-Nemo-Instruct-2407 Mistral-7B-Instruct-v0.37.2B · Dense · 2024-05-22GGUF Q4_K_M (bartowski) · 5.2 GiB–5.7 GiBmistralai/Mistral-7B-Instruct-v0.3 Mixtral-8x7B-Instruct-v0.147B · 13B active · MoE · 2023-12-10GGUF Q4_K_M (mradermacher) · 27.6 GiB–28.6 GiBmistralai/Mixtral-8x7B-Instruct-v0.1

Phi 4

Phi-4-reasoning15B · Dense · 2025-04-09GGUF Q4_K_M (Unsloth) · 10.1 GiB–10.8 GiBmicrosoft/Phi-4-reasoning Phi-4-mini-instruct3.8B · Dense · 2025-02-19GGUF Q4_K_M (Unsloth) · 3.4 GiB–3.9 GiBmicrosoft/Phi-4-mini-instruct phi-415B · Dense · 2024-12-11GGUF Q4_K_M (Unsloth) · 10.0 GiB–10.6 GiBmicrosoft/phi-4 Phi-3.5-mini-instruct3.8B · Dense · 2024-08-16GGUF Q4_K_M (bartowski) · 5.4 GiB–6.0 GiBmicrosoft/Phi-3.5-mini-instruct

Kimi 2

Kimi-K2-Thinking1026B · MoE · 2025-11-04GGUF Q4_K_M (Unsloth) · 579.2 GiB–680.0 GiBmoonshotai/Kimi-K2-Thinking Kimi-K2-Instruct1026B · MoE · 2025-07-11GGUF Q4_K_M (Unsloth) · 578.8 GiB–679.5 GiBmoonshotai/Kimi-K2-Instruct

MiniMax 2

MiniMax-M3427B · MoE · 2026-06-02GGUF Q4_K_M (bartowski) · 244.4 GiB–249.7 GiBMiniMaxAI/MiniMax-M3 MiniMax-M2229B · MoE · 2025-10-22GGUF Q4_K_M (Unsloth) · 130.9 GiB–134.0 GiBMiniMaxAI/MiniMax-M2

OpenAI 2

gpt-oss-120b117B · 5.1B active · MoE · 2025-08-04MXFP4 safetensors · 61.1 GiB–61.7 GiBopenai/gpt-oss-120b gpt-oss-20b21B · 3.6B active · MoE · 2025-08-04MXFP4 safetensors · 13.1 GiB–13.6 GiBopenai/gpt-oss-20b

Ant Group / inclusionAI 1

Ling-3.1-flash572B · 25B active · MoE · 2026-10-13Official FP8 safetensors (F8_E4M3 experts + BF16 remainder) · 535.0 GiB–616.1 GiBinclusionAI/Ling-3.1-flash

Fish Audio 1

Drama 336B · 3.0B active · MoE · 2026-10-13Unofficial int4 W4A16 (GPTQ-Marlin, with codec) · 22.6 GiB–26.6 GiBfishaudio/drama-3

IBM 1

granite-3.3-8b-instruct8.2B · Dense · 2025-04-09GGUF Q4_K_M (IBM) · 6.0 GiB–6.5 GiBibm-granite/granite-3.3-8b-instruct

Meta 1

Muse Spark Open822B · 42B active · MoE · 2026-10-15Official NVFP4 safetensors · 498.4 GiB–499.0 GiBmeta-models/Muse-Spark-Open-816B-A42B

Recent releases that are not in the calculator

We checked these releases on 16 October 2026. Some are closed or API-only, and the open ones use architectures the calculator does not model yet. Each note covers what the publisher has released: weights, file sizes, or only an API price.

FLUX 3 Image Klein: open weights, file sizes and the publisher's hardware notes.
Nemotron 3.5 Bolt 74B-A9B 24GB: open weights packed for a 24 GB card.
Nemotron 3.5 Super VL: open weights and NVIDIA's deployment targets.
Step 5 Preview: open weights and StepFun's multi-GPU examples.
Muse Spark Open: open weights, file sizes and Meta's two-machine note.
Muse Wonder: API price only. No weights or memory figures are published.
Gemini 4 Argon: API price only. No weights or memory figures are published.
MAI-Base-2: parameter counts from Microsoft. No public download and no price.