AI Model Hardware Requirements

What hardware do you actually need to run today's most popular AI models locally? We break down VRAM requirements, recommended GPUs, and compatible tools for each model — from lightweight 7B chatbots to 405B research monsters.

Llama 4

8B5 GB – 16 GB VRAM

Llama 4 Scout 8B

General-purpose chat, coding assistance, instruction following

View hardware picks
70B40 GB – 140 GB VRAM

Llama 4 Maverick 70B

Advanced reasoning, long-context analysis, complex coding tasks

View hardware picks
405B230 GB – 810 GB VRAM

Llama 4 Behemoth 405B

Research-grade reasoning, multi-step problem solving (multi-GPU required)

View hardware picks

DeepSeek R1

7B4.5 GB – 14 GB VRAM

DeepSeek R1 7B

Reasoning-focused chat, math and logic, chain-of-thought inference

View hardware picks
70B40 GB – 140 GB VRAM

DeepSeek R1 70B

Advanced reasoning, research-grade chain-of-thought, complex math

View hardware picks

Gemma 3

9B6 GB – 18 GB VRAM

Gemma 3 9B

Multilingual chat, summarization, lightweight instruction following

View hardware picks
27B16 GB – 54 GB VRAM

Gemma 3 27B

High-quality multilingual generation, reasoning, coding

View hardware picks

Qwen 3

7B4.5 GB – 14 GB VRAM

Qwen 3 7B

Multilingual chat, coding, tool use, function calling

View hardware picks
72B42 GB – 144 GB VRAM

Qwen 3 72B

Enterprise-grade multilingual generation, complex reasoning, agentic workflows

View hardware picks

Stable Diffusion

3.5B4 GB – 8 GB VRAM

Stable Diffusion XL

High-resolution image generation (1024x1024), artistic styles, inpainting

View hardware picks

Flux

12B8 GB – 24 GB VRAM

Flux.1 Dev

State-of-the-art text-to-image generation, photorealistic output, prompt adherence

View hardware picks

Whisper

1.5B1.5 GB – 4 GB VRAM

Whisper Large V3

Speech-to-text transcription, multilingual audio, real-time captioning

View hardware picks

Mistral

7B4.5 GB – 14 GB VRAM

Mistral 7B

Efficient general-purpose chat, summarization, instruction following

View hardware picks

Phi

14B9 GB – 28 GB VRAM

Phi-4 14B

Reasoning, math, coding — strong performance for its size class

View hardware picks

CodeLlama

34B20 GB – 68 GB VRAM

CodeLlama 34B

Code generation, code completion, debugging, technical documentation

View hardware picks