Skip to content
KoboldCpp
GitHub

Choose a model and launch

KoboldCpp doesn't come with an AI model; you add one. A model is one large file that ends in .gguf.

The easy way: a newbie template

Section titled “The easy way: a newbie template”

A template fills in the launcher for you and downloads the model when you click Launch. The Chatbot templates also add voice input, read-aloud and picture understanding.

  1. In the launcher, click Get Help and the dropdown Newbie Templates.

  2. Pick a template for your graphics card's memory (VRAM; how to check):

    • Chat with the AI:
      • LowSpec-Chatbot: recommend for 6 GB VRAM or no GPU. Chat or roleplay, also has vision, speech generation and voice recognition capabilities.
      • MidSpec-Chatbot: recommend for 12 GB VRAM or more. Chat or roleplay, also has vision, speech generation and voice recognition capabilities.
    • Generating Images:
      • LowSpec-ImageGen: recommend for 6 GB VRAM or no GPU for creating and editing AI images. Launch the dedicated SDUI when prompted.
      • LowSpec-ImageGen: recommend for 12 GB VRAM for creating and editing AI images. Launch the dedicated SDUI when prompted.
    • Generating Music:
      • LowSpec-MusicGen: recommend for 6 GB VRAM or no GPU for generating music. Launch the dedicated MusicUI when prompted. You can add an extra LLM model for better lyrics generation.
    • Agent and Coding:
      • LowSpec-CodingAgent: recommend for 6 GB VRAM or no GPU for coding tasks. Launches an agentic terminal.
      • MidSpec-CodingAgent: recommend for 12 GB VRAM for coding tasks. Launches an agentic terminal.

  3. Click Load Template.

    The Help Menu: Newbie Templates, the LowSpec-Chatbot template and the Load Template button

Or: download one model file

Section titled “Or: download one model file”

These are some basic models that work in KoboldCpp. Click one to download it. For more models, visit Huggingface.

Qwen3-VL-8BThe best all-rounder · 4.8 GB
L3-8B-Stheno-v3.2Creative writing and roleplay · 4.7 GB
Gemma 3 4BSmall and fast, for weaker PCs · 2.5 GB
PicX RealMakes pictures · 1.7 GB

In the launcher, click Browse next to GGUF Text Model: and select the file. PicX Real is an image model: it goes into Image Model on the Image Gen tab instead.

Launch

Section titled “Launch”

Click Launch.

GGUF Text Model field with a model selected (1) and the Launch button (2)

The launcher closes, and the console window shows the model loading. With a template, it first downloads the files, which can take a while. The next launch reuses them.