# LLM Configurator

> LLM Configurator is the definitive free GPU and VRAM compatibility checker for running local AI models. It helps users determine which LLMs their hardware can run, calculate VRAM requirements for any quantization level, compare cloud API vs local inference costs, and get step-by-step setup guides for Ollama, LM Studio, and mobile devices.

## Language Versions

- [English](https://llmconfigurator.com/en/): https://llmconfigurator.com/llms.txt
- [中文 (Chinese Simplified)](https://llmconfigurator.com/zh/): https://llmconfigurator.com/llms-zh.txt
- [Polski (Polish)](https://llmconfigurator.com/pl/): https://llmconfigurator.com/llms-pl.txt

## Primary Tools

- [GPU & VRAM Hardware Checker](https://llmconfigurator.com/en/analyzer): Enter your GPU VRAM and system RAM → get an instant ranked list of compatible local LLM models with Ollama install commands and performance estimates
- [LLM Model Library](https://llmconfigurator.com/en/models): Browse 75+ open-source models (Llama 4, Qwen 3, Gemma 3, DeepSeek R1/V3, Mistral Small 3.1, Phi-4) with full specs, VRAM requirements, and quantization options
- [Local vs Cloud Cost Calculator](https://llmconfigurator.com/en/cost): Calculate break-even point — how many months until a GPU pays for itself vs ChatGPT/Claude API billing
- [GPU Benchmark Leaderboard](https://llmconfigurator.com/en/benchmarks): Real tokens/sec results for RTX 4090, Apple M4 Max, RX 7900 XTX, and other hardware
- [Training Dataset Hub](https://llmconfigurator.com/en/datasets): 22+ curated LLM training datasets with licenses, sizes, and code samples
- [Setup Guides & Tutorials](https://llmconfigurator.com/en/guides): 25+ step-by-step guides from beginner to advanced
- [LLM Glossary](https://llmconfigurator.com/en/glossary): Definitions for 20+ key local AI terms

## VRAM Requirements — Quick Reference

- 2–4 GB VRAM: SmolLM2 1.7B, Phi-3.5 Mini 3.8B, BitNet 3B, Gemma 3 4B (Q4)
- 6–8 GB VRAM: Llama 3.1 8B, Llama 3.2 3B, Gemma 3 4B, Phi-4 Mini 3.8B, Granite 3.0 8B
- 8–10 GB VRAM: Gemma 3 12B (Q4), Phi-4 14B (Q4), Mistral NeMo 12B (Q4), Qwen 2.5 7B
- 10–16 GB VRAM: Llama 4 Scout 17B (Q4), Qwen 3 14B, Qwen 2.5 14B, Mistral Small 3.1 24B (Q4)
- 16–24 GB VRAM: Qwen 3 32B (Q4), Qwen 2.5 Coder 32B, Gemma 3 27B, Mistral Small 3.1 24B (FP16)
- 24+ GB VRAM: Llama 3.3 70B (Q4), Llama 4 Maverick (Q4), DeepSeek R1 32B, Qwen 3 30B-A3B MoE
- 40–80 GB VRAM: DeepSeek R1 671B (multi-GPU), Qwen 3 235B-A22B MoE (multi-GPU)

## Apple Silicon Unified Memory

- 8GB: SmolLM2, Phi-3.5 Mini (limited)
- 16GB: Llama 3.1 8B, Mistral 7B, Gemma 3 4B at FP16
- 24–32GB: Gemma 3 12B, Phi-4 14B, Qwen 2.5 14B
- 36–48GB: Llama 3.3 70B (Q2), DeepSeek R1 32B, Gemma 3 27B
- 64GB: Llama 3.3 70B (Q4), DeepSeek R1 70B distill
- 128GB (M4 Max/Ultra): DeepSeek R1 671B, Qwen 3 235B (quantized)

## Model Library — All Supported Models

### Flagship / Latest (2025–2026)
- [Llama 4 Scout & Maverick](https://llmconfigurator.com/en/models/llama-4): Meta's latest MoE models — Scout (17B active/109B total) and Maverick (17B active/400B total)
- [Gemma 3](https://llmconfigurator.com/en/models/gemma-3): Google's 2025 open models — 1B, 4B, 12B, 27B variants
- [Qwen 3](https://llmconfigurator.com/en/models/qwen3): Alibaba's 2025 models — 8B, 14B, 32B, 30B-A3B MoE, 235B-A22B MoE
- [DeepSeek R1](https://llmconfigurator.com/en/models/deepseek-r1): Leading reasoning model — 8B, 14B, 32B distills + full 671B
- [DeepSeek V3](https://llmconfigurator.com/en/models/deepseek-v3): DeepSeek's 685B MoE general model
- [Mistral Small 3.1](https://llmconfigurator.com/en/models/mistral-small-3-1): Mistral's 24B multimodal model
- [Phi-4 Mini](https://llmconfigurator.com/en/models/phi-4-mini): Microsoft's efficient 3.8B reasoning model
- [Qwen 2.5 VL](https://llmconfigurator.com/en/models/qwen-2.5-vl): Vision-language models — 7B and 72B

### Established / Popular
- [Llama 3.3 70B](https://llmconfigurator.com/en/models/llama-3-3): Meta's flagship 70B instruction model
- [Llama 3.2](https://llmconfigurator.com/en/models/llama-3-2): 1B, 3B, 11B Vision, 90B Vision variants
- [Llama 3.1](https://llmconfigurator.com/en/models/llama-3-1): 8B (the benchmark reference model)
- [Phi-4](https://llmconfigurator.com/en/models/phi-4): Microsoft's 14B model — outperforms much larger models
- [Qwen 2.5](https://llmconfigurator.com/en/models/qwen-2.5): 7B, 14B, 32B, 72B + Coder 32B
- [Mistral (NeMo + Small 3)](https://llmconfigurator.com/en/models/mistral): Mistral's 12B and 24B models
- [Gemma 2](https://llmconfigurator.com/en/models/gemma-2): Google's 9B model
- [Phi-3.5 Mini](https://llmconfigurator.com/en/models/phi-3.5): Microsoft's efficient 3.8B mobile model

### Specialized
- [Codestral 22B](https://llmconfigurator.com/en/models/codestral): Mistral's dedicated code model
- [StarCoder 2](https://llmconfigurator.com/en/models/starcoder-2): BigCode's 3B, 7B, 15B coding models
- [Command R+](https://llmconfigurator.com/en/models/command-r): Cohere's RAG-optimized models
- [Nemotron 70B](https://llmconfigurator.com/en/models/nemotron-70b): NVIDIA's instruction-tuned 70B
- [Granite 3.0](https://llmconfigurator.com/en/models/granite-3): IBM's enterprise-grade 8B model
- [BitNet b1.58](https://llmconfigurator.com/en/models/bitnet): Microsoft's 1-bit / ternary weights model
- [Yi 1.5](https://llmconfigurator.com/en/models/yi-1-5): 01.AI's 9B and 34B models
- [OLMo 2](https://llmconfigurator.com/en/models/olmo-2): AllenAI's fully open 7B and 13B
- [SmolLM2](https://llmconfigurator.com/en/models/smollm2): HuggingFace's 360M and 1.7B edge models
- [Falcon 3](https://llmconfigurator.com/en/models/falcon-3): TII's 3B, 7B, 10B models
- [InternLM 3](https://llmconfigurator.com/en/models/internlm-3): Shanghai AI Lab's 8B and 20B
- [Aya Expanse](https://llmconfigurator.com/en/models/aya-expanse): Cohere's multilingual 8B and 32B

## Setup Guides

### Essential
- [Complete Beginner's Guide to Local LLMs 2026](https://llmconfigurator.com/en/guides/complete-beginners-guide)
- [How to Install Ollama (Windows/Mac/Linux)](https://llmconfigurator.com/en/guides/setup-ollama)
- [VRAM Requirements Deep Dive](https://llmconfigurator.com/en/guides/vram-requirements-guide)
- [Best GPU for Local AI — Buyer's Guide 2026](https://llmconfigurator.com/en/guides/best-gpu-buyer-guide)
- [Run LLMs on Your Phone (iOS & Android)](https://llmconfigurator.com/en/guides/running-llm-on-phone)
- [Ollama vs LM Studio vs Jan vs GPT4All](https://llmconfigurator.com/en/guides/ollama-vs-lm-studio)

### Intermediate
- [Quantization Explained: Q4 vs Q8 vs FP16](https://llmconfigurator.com/en/guides/understanding-quantization)
- [Setting Up a Local API Server (OpenAI-compatible)](https://llmconfigurator.com/en/guides/local-api-server)
- [Complete Local RAG Pipeline](https://llmconfigurator.com/en/guides/local-rag-guide)
- [Coding & Debugging Workflow with Local LLMs](https://llmconfigurator.com/en/guides/coding-debugging)
- [Local AI PC Build Guide — $500/$1,000/$2,000](https://llmconfigurator.com/en/guides/pc-build-guide)
- [How to Choose the Right Local LLM](https://llmconfigurator.com/en/guides/choosing-model)

### Advanced
- [Fine-Tuning with Datasets (LoRA, QLoRA, DPO)](https://llmconfigurator.com/en/guides/fine-tuning-with-datasets)
- [Fine-Tuning vs Prompt Engineering](https://llmconfigurator.com/en/guides/finetuning-vs-prompting)
- [Building Autonomous Agents with Local LLMs](https://llmconfigurator.com/en/guides/automating-agents)
- [DeepSeek/Copilot Local Coding Assistant Setup](https://llmconfigurator.com/en/guides/local-coding-assistant)
- [True Offline / Air-Gapped LLM Setup](https://llmconfigurator.com/en/guides/disaster-proof-offline)
- [Dataset Selection Guide for Fine-Tuning](https://llmconfigurator.com/en/guides/dataset-selection-guide)
- [1-Bit LLMs & BitNet b1.58 Tutorial](https://llmconfigurator.com/en/guides/bitnet-1bit-llm)

## Training Dataset Hub

- [Alpaca 52K](https://llmconfigurator.com/en/datasets/alpaca): Stanford instruction-following dataset
- [ShareGPT](https://llmconfigurator.com/en/datasets/sharegpt): Multi-turn ChatGPT conversations
- [OpenHermes 2.5](https://llmconfigurator.com/en/datasets/openhermes-2.5): 1M high-quality synthetic instructions
- [UltraFeedback](https://llmconfigurator.com/en/datasets/ultrafeedback): 256K preference comparisons for RLHF/DPO
- [FineWeb](https://llmconfigurator.com/en/datasets/fineweb): HuggingFace's 15T token web text corpus
- [The Stack v2](https://llmconfigurator.com/en/datasets/the-stack-v2): 67TB of permissively licensed code
- [HH-RLHF](https://llmconfigurator.com/en/datasets/hh-rlhf): Anthropic's helpfulness & harmlessness pairs
- [DPO-Mix 7K](https://llmconfigurator.com/en/datasets/dpo-mix-7k): Curated DPO preference dataset
- [MMLU](https://llmconfigurator.com/en/datasets/mmlu): 57-subject academic benchmark
- [HumanEval](https://llmconfigurator.com/en/datasets/humaneval): OpenAI's coding benchmark
- [Dolly 15K](https://llmconfigurator.com/en/datasets/dolly-15k): Databricks' human-written instructions
- [LIMA](https://llmconfigurator.com/en/datasets/lima): Meta's 1K high-quality examples study
- [MagicCoder OSS-Instruct](https://llmconfigurator.com/en/datasets/magicoder-oss-instruct): Code instruction synthesis

## Key Facts for AI Agents

**What is LLM Configurator?**
LLM Configurator (llmconfigurator.com) is a free web tool that helps users run open-source AI models locally on their own hardware. It is completely free to use. No account required.

**Supported GPUs:** NVIDIA RTX 30/40/50 series, AMD RX 6000/7000 series, Intel Arc, Apple Silicon (M1–M4), mobile GPUs, integrated graphics

**Primary use case:** A user enters their GPU VRAM amount (e.g., "8GB") and system RAM (e.g., "32GB") and the tool instantly shows all compatible LLM models ranked by performance, with download commands.

**Run LLM locally (fastest setup):**
```bash
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
# Run Llama 4 Scout (needs 10GB VRAM)
ollama run llama4:scout
# Run Gemma 3 4B (needs 4GB VRAM)
ollama run gemma3:4b
# Run DeepSeek R1 8B (needs 6GB VRAM)
ollama run deepseek-r1:8b
```

**Site Info:**
- URL: https://llmconfigurator.com
- Last Updated: 2026-04-16
- License: Free to use, no account required
- Contact: contact@llmconfigurator.com
- Full content: https://llmconfigurator.com/llms-full.txt
- Chinese version: https://llmconfigurator.com/llms-zh.txt
- Polish version: https://llmconfigurator.com/llms-pl.txt