# CanItRun

> CanItRun is a free web calculator that tells you whether a specific open-weight large language model will run on your GPU, at which quantization level, and roughly how many tokens per second to expect. It also lists benchmark scores so you can compare quality.

CanItRun is a free web calculator that tells you whether a specific open-weight large language model will run on your GPU, at which quantization level, and roughly how many tokens per second to expect. It also lists benchmark scores so you can compare quality.

## Featured

- [DeepSeek R1](https://canitrun.dev/models/deepseek-r1/) -- top open reasoning model beating GPT-4o in math and coding
- [Llama 3.3 70B](https://canitrun.dev/models/llama-3.3-70b/) -- Meta's flagship open-weight LLM with 128K context
- [NVIDIA RTX 5090](https://canitrun.dev/gpus/rtx-5090/) -- latest Blackwell flagship with 32 GB GDDR7
- [NVIDIA RTX 4090](https://canitrun.dev/gpus/rtx-4090/) -- previous-gen flagship still widely used for local LLMs
- [GPU Comparisons](https://canitrun.dev/comparisons/) -- side-by-side GPU benchmarks and model compatibility

## Core pages

- [Calculator](https://canitrun.dev/): interactive hardware picker + model list.
- [All models](https://canitrun.dev/models/): every tracked LLM with params, context, license, benchmarks.
- [All GPUs](https://canitrun.dev/gpus/): every tracked GPU/SoC with VRAM, bandwidth, compatible models.
- [Apps & Agents](https://canitrun.dev/apps/): 20 AI apps with hardware requirements, model recommendations, and OpenRouter compatibility.
- [Methodology](https://canitrun.dev/about/): VRAM/KV-cache formulas and data sources.

## Apps & Agents

- [All apps](https://canitrun.dev/apps/): complete directory with search, filters, and hardware compatibility.
- [Coding Agents](https://canitrun.dev/apps/category/coding-agents/): Cline, Roo Code, Kilo Code, Aider, Continue, Claude Code, Cursor, Zed.
- [Chat Frontends](https://canitrun.dev/apps/category/chat-frontends/): Open WebUI, SillyTavern, LibreChat, Janitor AI, LM Studio, text-generation-webui, HuggingChat.
- [Local LLM Tools](https://canitrun.dev/apps/category/local-llm-tools/): LM Studio, Ollama, AnythingLLM, KoboldCPP, text-generation-webui, vLLM, llama.cpp.
- [OpenRouter Compatible](https://canitrun.dev/apps/category/openrouter-compatible/): 11 apps that support OpenRouter's unified API.
- [Self-Hosted Apps](https://canitrun.dev/apps/category/self-hosted/): Open WebUI, LibreChat, Ollama, vLLM, AnythingLLM.

### App comparisons

- [Cline vs Roo Code](https://canitrun.dev/apps/compare/cline-vs-roo-code/)
- [Open WebUI vs LM Studio](https://canitrun.dev/apps/compare/open-webui-vs-lm-studio/)
- [Ollama vs LM Studio](https://canitrun.dev/apps/compare/ollama-vs-lm-studio/)
- [Cline vs Aider](https://canitrun.dev/apps/compare/cline-vs-aider/)
- [Cline vs Continue](https://canitrun.dev/apps/compare/cline-vs-continue/)
- [Open WebUI vs LibreChat](https://canitrun.dev/apps/compare/open-webui-vs-librechat/)
- [SillyTavern vs Janitor AI](https://canitrun.dev/apps/compare/sillytavern-vs-janitor-ai/)
- [Claude Code vs Cursor](https://canitrun.dev/apps/compare/claude-code-vs-cursor/)
- [Ollama vs vLLM](https://canitrun.dev/apps/compare/ollama-vs-vllm/)
- [KoboldCPP vs text-generation-webui](https://canitrun.dev/apps/compare/koboldcpp-vs-text-generation-webui/)
- [Aider vs Continue](https://canitrun.dev/apps/compare/aider-vs-continue/)

## Guides & Tutorials

- [How Much VRAM Does Llama 3 Need?](https://canitrun.dev/guides/how-much-vram-for-llama-3/): VRAM requirements for all Llama 3 variants at every quantization level.
- [Best LLMs for 8 GB VRAM](https://canitrun.dev/guides/best-llms-for-8gb-vram/): which models fit on budget GPUs like the RTX 4060.
- [Best LLMs for 16 GB VRAM](https://canitrun.dev/guides/best-llms-for-16gb-vram/): the sweet spot tier for serious local inference.
- [How to Run LLMs on Mac](https://canitrun.dev/guides/run-llms-on-mac/): M1-M4 guide covering MLX, Ollama, and unified memory.
- [GGUF Quantization Explained](https://canitrun.dev/guides/quantization-explained/): Q4, Q5, Q6, Q8 compared with quality benchmarks.
- [Best GPU for Local LLMs](https://canitrun.dev/guides/best-gpu-for-local-llm/): buyer's guide covering NVIDIA, AMD, and Apple Silicon.
- [Getting Started with Ollama](https://canitrun.dev/guides/ollama-getting-started/): install, run, and customize models in one command.
- [GGUF vs EXL2 vs AWQ](https://canitrun.dev/guides/gguf-vs-exl2-vs-awq/): which quantization format to use and when.

## Data sources

Model metadata: [Hugging Face model cards](https://huggingface.co/). Quantized sizes: [Ollama library](https://ollama.com/library) GGUF builds.
Benchmarks: [Open LLM Leaderboard v2](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) (MMLU-Pro, GPQA, IFEval, MATH, BBH, MuSR), [LMSYS Chatbot Arena](https://lmarena.ai/) for ELO.
GPU specs: manufacturer datasheets and [TechPowerUp](https://www.techpowerup.com/gpu-specs/).

> Last updated: 2026-06-28