AI Model Hardware Requirements
What hardware do you actually need to run today's most popular AI models locally? We estimate the memory each model needs and match it to the unified-memory boxes that can run it — from lightweight 4B chatbots to flagships too big for any single box.
Llama
Llama 3.3 70B Instruct
Meta's flagship dense chat model — strong general reasoning, coding, and instruction-following that rivals much larger models. The default pick when a single big box has the memory to spare.
Llama 4 Scout
A mixture-of-experts model — only a fraction of its 109B total parameters are active per token, so it loads like a large model but runs far faster than its size suggests. A natural fit for unified-memory boxes with a long context window.
Llama 3.1 405B
Meta's largest dense model — 405B parameters that exceed any local box's memory even at Q4. Reach for Llama 3.3 70B locally instead.
Qwen3
Qwen3 8B
A capable small chat-and-reasoning model that fits comfortably even on budget boxes — a good everyday assistant when you don't need frontier-level depth.
Qwen3 14B
The mid-size Qwen3 — noticeably stronger reasoning and coding than the 8B while still fitting mid-range hardware at Q4.
Qwen3 32B
Qwen3's large dense model — near-flagship quality that still runs on a single unified-memory box, a strong local alternative to 70B-class models at lower memory cost.
DeepSeek-R1
DeepSeek-R1 Distill Qwen 7B
R1's chain-of-thought reasoning distilled into a 7B Qwen base — the lightest way to get R1-style step-by-step reasoning locally, small enough for budget boxes.
DeepSeek-R1 Distill Qwen 14B
The 14B R1 distill — a good balance of reasoning depth and hardware cost, running comfortably on mid-range unified-memory boxes.
DeepSeek-R1 Distill Qwen 32B
The strongest Qwen-based R1 distill — heavy reasoning that still fits a single large-memory box, the sweet spot for local R1-style work.
DeepSeek-R1 Distill Llama 70B
R1's reasoning distilled onto a Llama 70B base — the highest-quality distill, and the one to reach for when the full 671B R1 won't fit (it never does locally).
DeepSeek R1 671B
The full DeepSeek R1 — a 671B-total MoE that needs hundreds of gigabytes even at Q4, far beyond any single unified-memory box. Run a DeepSeek-R1 distill instead.
Gemma 3
Gemma 3 4B
Google's smallest current Gemma — tiny memory footprint and multimodal, ideal for the most memory-constrained budget boxes and always-on assistants.
Gemma 3 12B
The mid-size Gemma 3 — solid general-purpose quality with a long context window, fitting mid-range hardware at Q4.
Gemma 3 27B
The largest Gemma 3 — frontier-adjacent quality for a dense open model, comfortably runnable on a single large-memory box.
GPT-OSS
GPT-OSS 20B
OpenAI's small open-weight MoE — its 21B total parameters load like a mid-size model, but because only a fraction are active per token it runs fast on modest hardware.
GPT-OSS 120B
OpenAI's large open-weight MoE — its 117B total parameters fit a single high-memory box, and because only a fraction are active per token it runs far quicker than its total suggests.