Ollama Models Guide: Choose, Pull, and Run the Best Local LLMs

Ollama Models Guide: Choose, Pull, and Run the Best Local LLMs

Ollama makes it easy to run large language models on your own hardware. Choosing the right model matters: some are better at coding, others at chat, and smaller models run faster on laptops.

Popular Ollama models

| Model | Size | Best for | Notes |

| --- | --- | --- | --- |

| llama3.2 | 1B–3B | Fast chat, low latency | Great for laptops and edge devices |

| mistral | 7B | General reasoning, instruction following | Strong all-rounder |

| codellama | 7B–70B | Code generation and completion | Specialized for programming |

| deepseek-coder | 6.7B+ | Coding, math, long context | Excellent for technical tasks |

| phi4 | 14B | Reasoning, chat | Microsoft model, good efficiency |

| qwen2.5 | 7B–72B | Multilingual, coding | Alibaba model, strong benchmarks |

Pull a model

ollama pull mistral

Ollama downloads the model weights and stores them locally. You only need to pull once per machine.

Run a model interactively

ollama run mistral

Type prompts and get responses in the terminal. Press Ctrl+D to exit.

Run a model via API

curl http://localhost:11434/api/generate -d '
{
  "model": "mistral",
  "prompt": "Summarize the benefits of local LLMs."
}
'

List installed models

ollama list

Remove a model

ollama rm codellama

Choose by hardware

• Laptop with 8GB RAM: stick to 1B–3B models like llama3.2:1b
• Laptop with 16GB RAM: 7B models like mistral or llama3.2 run well
• Desktop with 32GB+ RAM: 13B–14B models for stronger reasoning
• Workstation with GPU: 70B+ models if you have the VRAM

Choose by task

• Coding: codellama, deepseek-coder, qwen2.5-coder
• Chat and brainstorming: mistral, llama3.2, phi4
• Long documents: models with larger context windows such as qwen2.5:32b
• Multilingual: qwen2.5, aya

Quantization and speed

Ollama serves quantized models by default. A model tag like mistral:7b-q4_0 uses 4-bit quantization, which is faster and smaller with minor quality loss. Use higher precision when accuracy is critical.

Create a custom model with a Modelfile

FROM mistral

SYSTEM """You are a helpful coding assistant. Answer concisely and include code examples."""

PARAMETER temperature 0.2

Save as Modelfile, then build and run:

ollama create my-coder -f Modelfile
ollama run my-coder

Next steps

• Combine models with the Ollama Python API for scripted workflows
• Build a local chat UI with Streamlit
• Use Ollama with LangChain or LlamaIndex for retrieval-augmented generation

Picking the right Ollama model lets you trade speed for capability and build local AI apps that match your hardware and use case.