# Ollama Model Library — Best AI Models for Local Use

> Curated guide to the best models available on Ollama for coding, chat, and reasoning. Compare Llama, Mistral, Gemma, Phi, and Qwen models for local AI development.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

```bash
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull and run the best coding model
ollama run qwen2.5-coder:7b

# Pull and run the best chat model
ollama run llama3.1:8b
```

## What is the Ollama Model Library?

Ollama hosts hundreds of open-source AI models ready for one-command local deployment. This guide covers the best models for different tasks — coding, chat, reasoning, and specialized use cases. All models run locally on your hardware with full privacy.

**Answer-Ready**: Ollama Model Library provides 500+ open-source AI models for local use. One-command download and run. Best coding models: Qwen2.5-Coder, DeepSeek-Coder. Best chat: Llama 3.1, Mistral. Best reasoning: Phi-3, Gemma 2. All run locally with full privacy.

**Best for**: Developers choosing the right local model for their use case. **Works with**: Ollama, Jan, Open WebUI, Claude Code (as backend). **Setup time**: Under 2 minutes per model.

## Best Models by Task

### Coding Models

| Model | Size | Strength | Command |
|-------|------|----------|---------|
| Qwen2.5-Coder | 7B/32B | Best overall coding | `ollama run qwen2.5-coder:7b` |
| DeepSeek-Coder V2 | 16B | Complex reasoning | `ollama run deepseek-coder-v2:16b` |
| CodeLlama | 7B/34B | Code completion | `ollama run codellama:7b` |
| Starcoder2 | 3B/7B/15B | Multi-language | `ollama run starcoder2:7b` |

### Chat Models

| Model | Size | Strength | Command |
|-------|------|----------|---------|
| Llama 3.1 | 8B/70B | Best general chat | `ollama run llama3.1:8b` |
| Mistral | 7B | Fast, efficient | `ollama run mistral` |
| Gemma 2 | 9B/27B | Google quality | `ollama run gemma2:9b` |
| Phi-3 | 3.8B/14B | Small but capable | `ollama run phi3:14b` |

### Reasoning Models

| Model | Size | Strength | Command |
|-------|------|----------|---------|
| Qwen2.5 | 7B/72B | Math & logic | `ollama run qwen2.5:7b` |
| Phi-3 Medium | 14B | Analytical tasks | `ollama run phi3:14b` |
| Mixtral | 8x7B | Expert mixture | `ollama run mixtral:8x7b` |

### Specialized Models

| Model | Use Case | Command |
|-------|----------|---------|
| LLaVA | Vision + text | `ollama run llava` |
| Nomic-Embed | Embeddings | `ollama run nomic-embed-text` |
| Whisper | Speech-to-text | Via whisper.cpp |

## Hardware Requirements

| Model Size | RAM Needed | GPU VRAM | Best For |
|-----------|------------|----------|----------|
| 3B | 4GB | 4GB | Laptops |
| 7B | 8GB | 8GB | Desktop |
| 13B | 16GB | 16GB | Workstation |
| 34B | 32GB | 24GB | Pro GPU |
| 70B | 64GB | 48GB | Server |

## Model Selection Guide

```
Need coding help?
  → Small project: qwen2.5-coder:7b
  → Complex code: deepseek-coder-v2:16b

Need general chat?
  → Best quality: llama3.1:8b (or 70b if you have the hardware)
  → Fastest: mistral:7b

Need reasoning?
  → Math/logic: qwen2.5:7b
  → Analysis: phi3:14b

Limited hardware?
  → phi3:3.8b or gemma2:2b
```

## FAQ

**Q: Which model is closest to GPT-4?**
A: Llama 3.1 70B or Qwen2.5 72B are the closest in general capability. For coding, Qwen2.5-Coder 32B rivals GPT-4 on benchmarks.

**Q: Can I use these with Claude Code?**
A: Yes, run Ollama as a local server and point Claude Code to `http://localhost:11434` as a custom endpoint.

**Q: How much disk space do models need?**
A: Roughly 1GB per billion parameters in Q4 quantization. A 7B model is ~4GB, 70B is ~40GB.

## Source & Thanks

> [Ollama Model Library](https://ollama.com/library) — 500+ models
>
> [ollama/ollama](https://github.com/ollama/ollama) — 120k+ stars

<!-- ZH -->

## 快速使用

```bash
ollama run llama3.1:8b
```

一行命令下载运行本地 AI 模型。

## Ollama 模型库指南

Ollama 提供 500+ 开源 AI 模型，一键本地部署。本指南覆盖编码、对话、推理最佳模型选择。

**一句话总结**：Ollama 模型库最佳选择指南，编码选 Qwen2.5-Coder，对话选 Llama 3.1，推理选 Phi-3，500+ 模型本地运行。

**适合人群**：选择本地 AI 模型的开发者。

## 最佳模型

### 编码：Qwen2.5-Coder 7B/32B
### 对话：Llama 3.1 8B/70B
### 推理：Qwen2.5 7B, Phi-3 14B

## 硬件要求
7B 模型需 8GB 内存，70B 需 64GB。

## 常见问题

**Q: 哪个最接近 GPT-4？**
A: Llama 3.1 70B 或 Qwen2.5 72B。

**Q: 能配合 Claude Code？**
A: 可以，Ollama 本地服务器作为自定义端点。

## 来源与致谢

> [ollama.com/library](https://ollama.com/library) — 500+ 模型, 120k+ stars

---
Source: https://tokrepo.com/en/workflows/4cecf968-aa84-47ec-9f32-c3b11432c18f
Author: Skill Factory