GPU Rental Prices
SearchCtrl K

LLM VRAM requirements

How much GPU memory each open model needs at 4 bits, with 4,096 tokens of context and 20% runtime headroom, and the cheapest card that holds it at the latest checked prices. Open a model to change precision, context and batch, or to size LoRA and full fine-tuning.

Work out a custom scenario in the LLM VRAM calculator, or check how much VRAM your own card has. Speech, image, video, OCR and embedding models are sized by their publishers' own figures.

74 language models in 10 families and 8 speech, image, video, OCR and embedding models · prices checked 21 November 2026

Llama

Qwen

DeepSeek

Mistral and Mixtral

Gemma

Phi

GPT-OSS

GLM

Kimi

Other open models

Speech, image, video, OCR and embedding models

Sized by their publishers' own figures, not the language-model estimate above.

All model memory estimates

ModelParameters4-bit weights, GBCache estimate
Kimi K2 09051026.41 billion513.2Not available
Kimi K21026.41 billion513.2Not available
DeepSeek V3.1684.53 billion342.3Not available
DeepSeek R1 0528684.53 billion342.3Not available
DeepSeek V3684.49 billion342.2Not available
DeepSeek R1684.49 billion342.2Not available
Qwen3 Coder 480B A35B480.15 billion240.1Config supported
Llama 3.1 405B405.85 billion202.9Not available
Llama 4 Maverick 17B 128E401.58 billion200.8Not available
GLM 4.5358.34 billion179.2Config supported
GLM 4.6356.79 billion178.4Config supported
Qwen3 235B A22B235.09 billion117.5Config supported
Qwen3 235B A22B 2507235.09 billion117.5Config supported
Mixtral 8x22B v0.1140.63 billion70.3Config supported
gpt-oss-120b117.00 billion58.5Not available
GLM 4.5 Air110.47 billion55.2Config supported
Llama 4 Scout 17B 16E108.64 billion54.3Not available
Qwen3 Next 80B A3B81.32 billion40.7Not available
Qwen2.5 72B72.71 billion36.4Config supported
Llama 3.1 70B70.55 billion35.3Not available
Llama 3.3 70B70.55 billion35.3Not available
DeepSeek R1 Distill Llama 70B70.55 billion35.3Config supported
Mixtral 8x7B v0.146.70 billion23.4Config supported
Qwen2.5 32B32.76 billion16.4Config supported
Qwen2.5 Coder 32B32.76 billion16.4Config supported
DeepSeek R1 Distill Qwen 32B32.76 billion16.4Config supported
QwQ 32B32.76 billion16.4Config supported
Qwen3 32B32.76 billion16.4Config supported
Qwen3 30B A3B30.53 billion15.3Config supported
Qwen3 30B A3B 250730.53 billion15.3Config supported
Qwen3 Coder 30B A3B30.53 billion15.3Config supported
gemma 3 27b27.43 billion13.7Not available
gemma 2 27b27.23 billion13.6Not available
Mistral Small 3.2 24B 250624.01 billion12.0Config supported
Mistral Small 24B 250123.57 billion11.8Config supported
Devstral Small 250723.57 billion11.8Config supported
gpt-oss-20b21.00 billion10.5Not available
Qwen2.5 14B14.77 billion7.4Config supported
Qwen2.5 Coder 14B14.77 billion7.4Config supported
DeepSeek R1 Distill Qwen 14B14.77 billion7.4Config supported
Qwen3 14B14.77 billion7.4Config supported
phi 414.66 billion7.3Config supported
Phi 4 reasoning14.66 billion7.3Config supported
Phi 4 reasoning plus14.66 billion7.3Config supported
Phi 3 medium 4k13.96 billion7.0Not available
Mistral Nemo 240712.25 billion6.1Config supported
gemma 3 12b12.19 billion6.1Not available
gemma 2 9b9.24 billion4.6Not available
Qwen3 8B8.19 billion4.1Config supported
Llama 3.1 8B8.03 billion4.0Not available
DeepSeek R1 Distill Llama 8B8.03 billion4.0Config supported
Ministral 8B 24108.02 billion4.0Not available
Qwen2.5 7B7.62 billion3.8Config supported
Qwen2.5 Coder 7B7.62 billion3.8Config supported
DeepSeek R1 Distill Qwen 7B7.62 billion3.8Config supported
Mistral 7B v0.37.25 billion3.6Config supported
gemma 3 4b4.30 billion2.2Not available
Qwen3 4B4.02 billion2.0Config supported
Qwen3 4B 25074.02 billion2.0Config supported
Phi 4 mini3.84 billion1.9Not available
Phi 3 mini 4k3.82 billion1.9Not available
Llama 3.2 3B3.21 billion1.6Not available
Qwen2.5 3B3.09 billion1.5Config supported
Qwen2.5 Coder 3B3.09 billion1.5Config supported
gemma 2 2b2.61 billion1.3Not available
Qwen3 1.7B2.03 billion1.0Config supported
DeepSeek R1 Distill Qwen 1.5B1.78 billion0.9Config supported
Qwen2.5 1.5B1.54 billion0.8Config supported
Qwen2.5 Coder 1.5B1.54 billion0.8Config supported
Llama 3.2 1B1.24 billion0.6Not available
gemma 3 1b1.00 billion0.5Not available
Qwen3 0.6B0.75 billion0.4Config supported
Qwen2.5 0.5B0.49 billion0.2Config supported
Qwen2.5 Coder 0.5B0.49 billion0.2Config supported

Estimate training cost · Compare LLM API pricing · GPU rental prices

Compare your offers

Saved prices keep their original dates. Confirm the configuration, terms and availability at the source.

Select up to three offers to compare their rates and terms.

Type a GPU, a provider, a model or a question.