Ollama

For

One command to a running model, then you choose the front end.

Install

• macOS

  macOS

  Copy

  curl -fsSL https://ollama.com/install.sh | sh

  Official install page

• Linux

  Linux

  Copy

  curl -fsSL https://ollama.com/install.sh | sh

  Official install page

• Windows

  Windows

  Copy

  irm https://ollama.com/install.ps1 | iex

  Paste this in PowerShell. The same page also links an .exe.

  Official install page

Latest release

v0.35.1 · Sep 29

Maintenance release, see changelog

Release notes

Related news

• Ollama publishes v0.35.1

  2026-09-29

• Ollama now supports Jev-style decision models

  2026-09-29

• Ollama now supports Jev-style decision models

  2026-09-29

• Ollama publishes v0.35.0

  2026-09-28

• Ollama v0.40.0 release candidate

  2026-09-25

Models it runs well

• DeepSeek coder 16B

  A desktop coder, not the V4 Spark card.

• DeepSeek R1-class 70B

  Distilled reasoning size. A Spark or a pair, not a 4090.

• DeepSeek V3 ~671B-class

  Four Sparks. Not the two-node V4 Flash card.

• Gemma 2 2B

  An older small Gemma. Still a floor, not a coder.

• Gemma 2 9B

  A 16 GB laptop can hold it. A 4090 is wasted on it.

• Gemma 3 12B

  The honest small-laptop ceiling for Gemma.

30 more are on the models index.