Skip to content
Vibecoding Guide

Open Source Models

Run a model

Section titled “Run a model”
Terminal window
ollama run llama3.2          # download and start chatting
ollama pull mistral          # download without running
ollama list                  # list installed models
ollama rm llama3.2           # remove a model

Model quick reference

Section titled “Model quick reference”
llama3.2        3B    Meta        General purpose, fast, runs on most laptops
llama3.1        8B    Meta        Stronger reasoning, needs ~16GB RAM
mistral         7B    Mistral     Fast, good instruction following
mixtral         47B   Mistral     Very capable, needs high-end hardware
qwen2.5-coder   7B    Alibaba     Best local option for code generation
phi4            14B   Microsoft   Strong performance for its size
gemma2          9B    Google      Efficient, good on Apple Silicon
deepseek-coder  6.7B  DeepSeek    Solid for code, fast

Hardware guide

Section titled “Hardware guide”
3B–4B model    →  8GB RAM      runs on most modern laptops
7B–8B model    →  16GB RAM     comfortable on M-series Mac or 16GB PC
13B model      →  32GB RAM     needs a well-specced machine
70B model      →  64GB+ RAM    desktop GPU or Apple Silicon Max/Ultra

Apple Silicon tip

Section titled “Apple Silicon tip”
Terminal window
# Ollama uses Metal GPU acceleration on Apple Silicon automatically.
# An M2 Pro (16GB) runs 7B models well.
# An M3 Max (36GB) runs 13B models comfortably.

Useful flags

Section titled “Useful flags”
Terminal window
ollama run mistral --verbose        # show token speed and model info
OLLAMA_NUM_PARALLEL=2 ollama serve  # allow 2 concurrent requests

Browse models

Section titled “Browse models”

Find more at: https://ollama.com/library