ModelCards

Ollama

A desktop app, CLI and local server that pulls ready-packaged open models from its own library and runs them with one command; aimed at developers working on their own machine.

GitHub ↗Docs ↗Supported models ↗Website ↗

Rows marked partial with a cloud tag have no downloadable tag in the Ollama library: they run on ollama.com (Ollama Cloud), not on your machine — https://docs.ollama.com/cloud.

You can also import your own GGUF via a Modelfile, but Ollama does not quantize GGUF models during import — https://docs.ollama.com/import.

At a glance

Built by
Ollama
Status
Active
Latest release
v0.40.7 (2026-10-18)
License
MIT
Written in
Go
Serves
Language, Embedding
Hardware
NVIDIA GPU, AMD GPU, Apple Silicon (Metal), Vulkan GPUs, CPU
API
REST API on localhost:11434, plus OpenAI- and Anthropic-compatible endpoints
Quantization
GGUF (e.g. q4_K_M, q8_0), BF16/FP16, MXFP8, NVFP4, MLX

From Ollama's own repo and docs, checked 21 Oct 2026.

Get started

Install
curl -fsSL https://ollama.com/install.sh | sh
Run a model
ollama run {model}

Replace {model} with the model ID below.

What it's good at

  • One-command models from a curated library

    Models are pulled by library tag (e.g. gemma4:31b); tags pick size and format, including q4_K_M, q8_0, bf16, mxfp8, nvfp4 and MLX builds. source ↗

  • Local and cloud from the same CLI

    Tags ending in cloud run on Ollama's servers with no download (sign-in required); cloud features can be turned off to stay local-only. source ↗

  • Familiar APIs

    A native REST API (chat, generate, embed) on port 11434, with OpenAI- and Anthropic-compatible endpoints for existing clients. source ↗

  • Broad consumer GPU support

    NVIDIA GPUs from compute capability 5.0 (driver 550+), AMD via ROCm v7, Apple GPUs via Metal, and other GPUs on Windows/Linux via Vulkan. source ↗

  • Built on llama.cpp and MLX

    The README names llama.cpp as its backend; library pages also offer MLX builds for Apple Silicon, and release notes update both. source ↗

Catalog models it runs 25

Only models with an official source are listed. Ollama may run others with the same architecture; see its supported-models page.