AI Basics6 min readPublished: 2026-07-28Updated: 2026-08-07

Ollama Models List & Local LLM Guide (2026)

Ollama lets you run large language models locally. Here's the list of popular Ollama models, how to install them, and how to use the local API endpoint.

By Heizi· Founder & Editor· Published: 2026-07-28· Updated: 2026-08-07

What is Ollama?

Ollama is an open-source tool that lets you run large language models locally on your own machine. It handles model downloading, quantization, and serving, making local LLM deployment as simple as running a single command.

Unlike cloud APIs, Ollama runs entirely on your hardware — no API key, no per-token billing, no data leaving your machine. This makes it ideal for privacy-sensitive applications, offline use, and cost-free experimentation. For a broader overview of AI tools and concepts, the beginner's guide to AI provides context on where local models fit in the ecosystem.

Popular Ollama Models List

Ollama supports a wide range of open-source models. Here are the most popular ones in 2026, with their parameter sizes and approximate memory requirements. The models range from lightweight options that run on laptops to powerful models that need dedicated GPUs.

llama3.1:8b — Meta's Llama 3.1, 8B params, ~5GB RAM
llama3.1:70b — Llama 3.1, 70B params, ~40GB RAM (requires strong GPU)
qwen2.5:7b — Alibaba's Qwen 2.5, 7B params, ~5GB RAM, strong Chinese support
deepseek-r1:7b — DeepSeek-R1 distilled reasoning model, 7B params, ~5GB RAM
deepseek-r1:14b — DeepSeek-R1, 14B params, ~9GB RAM, better reasoning
gemma2:9b — Google's Gemma 2, 9B params, ~6GB RAM
mistral:7b — Mistral 7B, lightweight and fast, ~5GB RAM
phi3:3.8b — Microsoft Phi-3 mini, 3.8B params, ~3GB RAM, great for low-end hardware

Choosing the Right Model for Your Hardware

Selecting the right model depends on your hardware capabilities and use case. For laptops and basic desktops with 8-16GB RAM, the 7-8B parameter models (Llama 3.1 8B, Qwen 2.5 7B, Mistral 7B) are ideal — they offer good quality at manageable memory footprints and generate text at reasonable speeds.

For machines with 16-32GB RAM or a mid-range GPU (8-16GB VRAM), the 9-14B models (Gemma 2 9B, DeepSeek-R1 14B) provide better reasoning and knowledge depth. These models are suitable for more complex tasks like code generation and detailed analysis.

For servers with high-end GPUs (40GB+ VRAM), the 70B models like Llama 3.1 70B deliver near-GPT-4 quality for many tasks. However, inference is significantly slower, and you should carefully evaluate whether the quality improvement justifies the hardware cost.

If you have limited hardware (4GB RAM), the Microsoft Phi-3 mini (3.8B) is an excellent choice. It runs on almost any machine and still handles basic conversation, Q&A, and simple coding tasks reasonably well.

How to Install and Pull Models

Install Ollama from ollama.com — there are installers for macOS, Linux, and Windows. After installation, pulling a model is a single command:

bash

# Pull the Llama 3.1 8B model
ollama pull llama3.1:8b

# Run the model interactively
ollama run llama3.1:8b

# List installed models
ollama list

Basic Ollama commands to pull, run, and list models

Using the Ollama Local API

Ollama runs a local HTTP server on port 11434. The API is OpenAI-compatible, so you can use the same SDK with the base URL set to localhost. No API key is required for local calls, which is a significant advantage over cloud APIs that require authentication credentials — you simply point your code at localhost and start generating.

python

from openai import OpenAI

# Ollama runs locally — no API key needed
client = OpenAI(
    api_key="ollama",  # any string works
    base_url="http://localhost:11434/v1"
)

response = client.chat.completions.create(
    model="llama3.1:8b",
    messages=[
        {"role": "user", "content": "Explain what an API key is."}
    ]
)

print(response.choices[0].message.content)

Calling a local Ollama model through the OpenAI-compatible API

Ollama vs Cloud APIs: When to Use Which

Ollama shines for privacy-sensitive work, offline environments, and cost-free unlimited usage. Cloud APIs offer more powerful models (GPT-4o, Claude 3.5 Sonnet) that can't run on consumer hardware, plus features like function calling and vision that local models may not fully support. If you're looking for free cloud-based options, our free AI API guide covers providers like Google Gemini and OpenRouter that offer free tiers.

A common pattern: prototype locally with Ollama, then deploy to cloud APIs for production. This gives you free development and the best models for production. For the cloud deployment phase, setting up an OpenRouter API key is a convenient option since one key gives you access to hundreds of models from different providers.

FAQ

Is Ollama completely free?

Yes. Ollama is open-source and free. You download and run models on your own hardware at no cost. The only cost is your own compute resources (CPU, RAM, GPU).

Do I need an API key for Ollama?

No. Ollama runs locally on your machine. The local API endpoint doesn't require authentication. When using the OpenAI SDK, you can pass any string as the API key — it's ignored.

What hardware do I need to run Ollama models?

For 7-8B parameter models, you need about 5GB RAM (8GB+ recommended). For 14B models, about 9GB. For 70B models, you need a strong GPU with 40GB+ VRAM. Smaller models like Phi-3 (3.8B) can run on machines with as little as 4GB RAM.

Can Ollama models be used in production?

Yes, but with caveats. Local deployment is great for privacy and cost control, but performance and model quality may lag behind top cloud models. For production use, ensure you have adequate hardware and monitor response latency.