Ollama Run Models

To run models in Ollama you use two commands: ollama pull to download a model and ollama run to start chatting with it. ollama run also pulls automatically if the model is missing.

Pull a Model

ollama pull gemma3
pulling manifest
pulling aeda25e63ebd: 100% ▕████████████████▏ 3.3 GB
verifying sha256 digest
writing manifest
success

Chat Interactively

Running a model without a prompt opens an interactive session. Type messages and press Enter; type /bye to quit.

ollama run gemma3
>>> What is the capital of Japan?
The capital of Japan is Tokyo.

>>> /bye
(Sample response.)

Model Names and Tags

A model name has the form name:tag. The tag usually picks a size or quantization. If you omit it, Ollama uses :latest.

ollama run qwen3:4b       # 4B parameter variant
ollama run qwen3:14b      # bigger, smarter, needs more memory
ollama run llama3.2:1b    # tiny model, runs anywhere
(each command downloads that exact variant on first use)

Popular Open-Weight Models

At the time of writing (2026) the Ollama library includes model families such as these. New versions appear frequently, so browse ollama.com/library before choosing.

  • Llama (Meta) - general-purpose chat models such as llama3.2 and llama3.1.
  • Gemma (Google) - gemma3, strong small models, some with image input.
  • Qwen (Alibaba) - qwen3 for chat and reasoning, qwen2.5-coder for code.
  • Mistral - mistral and related models from Mistral AI.
  • DeepSeek-R1 - reasoning models that "think" before answering.
  • gpt-oss - OpenAI open-weight models.
  • nomic-embed-text and similar - embedding models for search (not chat).

One-Shot Prompts and Piping

Pass the prompt as an argument to get one answer and exit. You can also pipe file contents into a model, which is handy in scripts.

ollama run llama3.2 "Summarise this:" < notes.txt
The notes describe a plan to migrate the billing service to a new database by Q3...
(Sample response.)

Frequently asked questions

What is the best Ollama model?

There is no single best one. For general chat a 7B-14B Llama, Gemma or Qwen model is a strong start; for code try a coder model; for search use an embedding model.

How do I exit ollama run?

Type /bye or press Ctrl+D in the interactive session.

What does :latest mean in Ollama?

It is the default tag a model publisher marks as recommended, usually a mid-size 4-bit quantized variant.