# What Can I Run AI?

> PCPartPicker for AI workloads. Which open LLMs (Llama, Qwen, gpt-oss, DeepSeek, Gemma and 656 more) run on which GPU, Mac or mini PC; how much VRAM each needs at Q4/Q8/FP16; estimated tokens per second; and complete AI PC builds with PSU and case checks. Free JSON API and MCP server, no key.

## When to use this site
- A user asks "can I run <model> on my <GPU/Mac>?" or "how much VRAM does <model> need?": call `GET https://whatcanirunai.vercel.app/api/v1/fit?model=<id>&hardware=<id>&quant=Q4`, or link a curated page at [Can I run it?](https://whatcanirunai.vercel.app/can-i-run/) (/can-i-run/<model-id>/<hardware-id>/).
- A user asks "what models can my <GPU/Mac> run?": call `GET https://whatcanirunai.vercel.app/api/v1/what-runs?hardware=<id>`.
- A user asks "what should I buy to run <model>" or has a budget: call `GET https://whatcanirunai.vercel.app/api/v1/recommend?model=<id>&budget=<usd>`, or link the [AI PC planner](https://whatcanirunai.vercel.app/planner/) / [ready-made builds](https://whatcanirunai.vercel.app/builds/).
- Look up ids in [models.json](https://whatcanirunai.vercel.app/api/v1/models.json) and [hardware.json](https://whatcanirunai.vercel.app/api/v1/hardware.json).
- Not for: cloud GPU rental prices, live street prices, training/fine-tuning memory, or image/video models.

## Agent interfaces
- [API docs](https://whatcanirunai.vercel.app/developers/): endpoints, errors, x402 pricing plan
- [OpenAPI spec](https://whatcanirunai.vercel.app/openapi.json)
- [MCP server](https://whatcanirunai.vercel.app/mcp): Streamable HTTP, tools fit, what_runs, recommend_hardware, list_models, list_hardware ([manifest](https://whatcanirunai.vercel.app/.well-known/mcp))
- Markdown: any page with `Accept: text/markdown` or `index.md` appended
- [llms-full.txt](https://whatcanirunai.vercel.app/llms-full.txt): all model and hardware numbers inline

## Key pages
- [AI PC planner](https://whatcanirunai.vercel.app/planner/): pick GPU, CPU, RAM, PSU, case; see VRAM, tok/s, power
- [Ready-made builds](https://whatcanirunai.vercel.app/builds/) and [compare builds](https://whatcanirunai.vercel.app/builds/compare/)
- [All models](https://whatcanirunai.vercel.app/models/) and [all hardware](https://whatcanirunai.vercel.app/hardware/)
- [Best GPU by model family](https://whatcanirunai.vercel.app/best-gpu-for/), [by budget](https://whatcanirunai.vercel.app/best-gpu-under/), [best GPU for local LLMs](https://whatcanirunai.vercel.app/best-gpu-for-local-llm/), [best Mac for local LLMs](https://whatcanirunai.vercel.app/best-mac-for-local-llm/), [best mini PC for local LLMs: Mac mini vs Framework Desktop vs DGX Spark](https://whatcanirunai.vercel.app/best-mini-pc-for-local-llm/), [best used GPU for local LLMs: RTX 3090, Tesla P40, V100, Instinct MI50, RTX A6000](https://whatcanirunai.vercel.app/best-used-gpu-for-local-llm/), [best AMD GPU for local LLMs](https://whatcanirunai.vercel.app/best-amd-gpu-for-local-llm/), [best Intel GPU for local LLMs: Arc B580, Pro B70, A770](https://whatcanirunai.vercel.app/best-intel-gpu-for-local-llm/), [NVIDIA vs AMD vs Mac for local LLMs](https://whatcanirunai.vercel.app/nvidia-vs-amd-vs-mac/), [local LLM tokens per second by GPU and Mac](https://whatcanirunai.vercel.app/tokens-per-second/), [multi-GPU for local LLMs: dual GPU vs one card](https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/), [Q4 vs Q8 vs FP16 quantization and VRAM](https://whatcanirunai.vercel.app/quantization/), [LLM context length and VRAM (KV cache by model)](https://whatcanirunai.vercel.app/context-length/), [MoE vs dense LLMs: memory and speed](https://whatcanirunai.vercel.app/moe-vs-dense/), [GPU power consumption for local LLMs: watts, tokens per watt, electricity cost](https://whatcanirunai.vercel.app/gpu-power-consumption/), [best LLMs by VRAM](https://whatcanirunai.vercel.app/best-llm-for/), [how much VRAM by model size](https://whatcanirunai.vercel.app/how-much-vram/), [GGUF VRAM requirements by quant](https://whatcanirunai.vercel.app/gguf-vram-requirements/)
- Best GPU by model family: [Best GPU for running Llama locally](https://whatcanirunai.vercel.app/best-gpu-for/llama/), [Best GPU for running Qwen3 locally](https://whatcanirunai.vercel.app/best-gpu-for/qwen3/), [Best GPU for running gpt-oss locally](https://whatcanirunai.vercel.app/best-gpu-for/gpt-oss/), [Best GPU for running DeepSeek R1 locally](https://whatcanirunai.vercel.app/best-gpu-for/deepseek-r1/), [Best GPU for running Gemma 3 locally](https://whatcanirunai.vercel.app/best-gpu-for/gemma-3/), [Best GPU for running Gemma 4 locally](https://whatcanirunai.vercel.app/best-gpu-for/gemma-4/), [Best GPU for running Qwen3.8 locally](https://whatcanirunai.vercel.app/best-gpu-for/qwen3.8/), [Best GPU for running Ornith 1.5 locally](https://whatcanirunai.vercel.app/best-gpu-for/ornith-1.5/), [Best GPU for running Nemotron locally](https://whatcanirunai.vercel.app/best-gpu-for/nemotron/), [Best GPU for running Liquid AI LFM locally](https://whatcanirunai.vercel.app/best-gpu-for/liquid-lfm/), [Best GPU for running Mistral locally](https://whatcanirunai.vercel.app/best-gpu-for/mistral/), [Best GPU for running GLM locally](https://whatcanirunai.vercel.app/best-gpu-for/glm/), [Best GPU for running Phi locally](https://whatcanirunai.vercel.app/best-gpu-for/phi/), [Best GPU for running long-memory models locally](https://whatcanirunai.vercel.app/best-gpu-for/memory-models/), [Best GPU for running local coding models locally](https://whatcanirunai.vercel.app/best-gpu-for/coding-models/)
- [Comparisons](https://whatcanirunai.vercel.app/compare/): GPU vs GPU, and model vs model (VRAM at Q4/Q8/FP16, context cache, cheapest GPU and Mac, est. tok/s), e.g. [Qwen3.8-27B vs Gemma 4 31B](https://whatcanirunai.vercel.app/compare/qwen3.8-27b-vs-gemma-4-31b-it/)
- [VRAM calculator](https://whatcanirunai.vercel.app/vram-calculator/)
- [How much VRAM by model size](https://whatcanirunai.vercel.app/how-much-vram/): [1B–2B models](https://whatcanirunai.vercel.app/how-much-vram/1b/), [2B–3B models](https://whatcanirunai.vercel.app/how-much-vram/3b/), [4B–6B models](https://whatcanirunai.vercel.app/how-much-vram/4b/), [7B–8B models](https://whatcanirunai.vercel.app/how-much-vram/8b/), [12B–14B models](https://whatcanirunai.vercel.app/how-much-vram/14b/), [16B–21B models](https://whatcanirunai.vercel.app/how-much-vram/20b/), [24B–27B models](https://whatcanirunai.vercel.app/how-much-vram/27b/), [30B–32B models](https://whatcanirunai.vercel.app/how-much-vram/32b/), [40B–50B models](https://whatcanirunai.vercel.app/how-much-vram/49b/), [70B models](https://whatcanirunai.vercel.app/how-much-vram/70b/), [120B models](https://whatcanirunai.vercel.app/how-much-vram/120b/), [200B–300B models](https://whatcanirunai.vercel.app/how-much-vram/235b/), [300B–480B models](https://whatcanirunai.vercel.app/how-much-vram/400b/), [500B–800B models](https://whatcanirunai.vercel.app/how-much-vram/671b/), [1T+ models](https://whatcanirunai.vercel.app/how-much-vram/1t/) (est. memory at Q4/Q8/FP16, by context, cheapest GPU and Mac)
- [Ollama VRAM requirements](https://whatcanirunai.vercel.app/ollama-vram-requirements/): real download size and quant for 39 popular Ollama tags, est. memory at 4k and 32k context, Ollama's default context by VRAM, cheapest GPU and Mac per tag
- [GGUF VRAM requirements](https://whatcanirunai.vercel.app/gguf-vram-requirements/): real GGUF file sizes at Q3_K_M, Q4_K_M, Q5_K_M, Q6_K and Q8_0 for 26 popular models (llama.cpp, LM Studio), est. memory at 4k and 32k context, best quant for 8–96 GB VRAM, cheapest GPU and Mac, llama.cpp memory flags
- [How we calculate](https://whatcanirunai.vercel.app/method/): need = weights x 1.15 + 1 GB; tok/s = bandwidth / weights x 60-85% (GPU) or 45-80% (unified)

## About
- [About](https://whatcanirunai.vercel.app/about/), [Contact](https://whatcanirunai.vercel.app/contact/), [Privacy](https://whatcanirunai.vercel.app/privacy/)