Skip to content

Ollama Model Recommendations

This guide helps you choose the right models for your development needs based on hardware, use case, and performance requirements.

Model Selection Matrix

Use Case Recommended Models Memory Best For
Quick Code Snippets codellama:7b 8GB Fast responses, simple tasks
Code Reviews codellama:13b, deepseek-coder:6.7b 16GB Thorough analysis, bug detection
Complex Development codellama:34b, deepseek-coder:33b 32GB+ Architecture, complex algorithms
Documentation llama2:13b, mistral:7b 16GB Technical writing, explanations
Learning & Research mixtral:8x7b, llama2:70b 64GB+ Deep explanations, comparisons

Code-Focused Models

CodeLlama Family (Meta)

CodeLlama 7B - Best for quick tasks 

ollama pull codellama:7b
# Size: 4.8GB | RAM: 8GB | Speed: ⭐⭐⭐⭐⭐

Strengths: - Fast inference times - Good at code completion - Minimal resource usage - Great for interactive development

Best for: - Simple function generation - Quick bug fixes - Code explanations - Learning programming concepts

Example Use: 

ollama run codellama:7b "Write a Python function to validate email addresses"

CodeLlama 13B - Balanced performance 

ollama pull codellama:13b
# Size: 7.6GB | RAM: 16GB | Speed: ⭐⭐⭐⭐

Strengths: - Better code understanding - More accurate suggestions - Good debugging capabilities - Handles medium complexity

Best for: - Code reviews - Refactoring suggestions - API design - Testing strategies

Example Use: 

ollama run codellama:13b "Review this FastAPI endpoint for security issues"

CodeLlama 34B - Professional grade 

ollama pull codellama:34b
# Size: 20GB | RAM: 32GB+ | Speed: ⭐⭐⭐

Strengths: - Advanced code analysis - Complex problem solving - Architectural guidance - Multi-language expertise

Best for: - System design discussions - Performance optimization - Complex debugging - Large codebase analysis

Example Use: 

ollama run codellama:34b "Design a scalable microservices architecture for e-commerce"

Specialized Versions

CodeLlama Instruct - Better at following instructions 

ollama pull codellama:7b-instruct
ollama pull codellama:13b-instruct

Use when: - Need precise output format - Following complex instructions - Step-by-step code generation - Structured responses required

CodeLlama Python - Python-specialized 

ollama pull codellama:7b-python

Use when: - Python-heavy projects - Django/Flask development - Data science workflows - Python best practices needed

Advanced Code Models

DeepSeek Coder

DeepSeek Coder 6.7B 

ollama pull deepseek-coder:6.7b
# Size: 4.2GB | RAM: 8GB | Speed: ⭐⭐⭐⭐⭐

Strengths: - Excellent code understanding - Strong debugging capabilities - Good at code optimization - Efficient resource usage

Best for: - Algorithm implementation - Performance optimization - Code quality improvement - Complex logic debugging

DeepSeek Coder 33B 

ollama pull deepseek-coder:33b
# Size: 19GB | RAM: 32GB+ | Speed: ⭐⭐⭐

Strengths: - Superior code analysis - Advanced problem solving - Multi-language expertise - Enterprise-level capabilities

Best for: - Large-scale refactoring - Architecture reviews - Security analysis - Complex system design

StarCoder

StarCoder 7B 

ollama pull starcoder:7b
# Size: 4.3GB | RAM: 8GB | Speed: ⭐⭐⭐⭐

Strengths: - Trained on permissive licenses - Good code completion - Multi-language support - Open-source friendly

Best for: - Open source projects - Code completion - Learning new languages - Collaborative development

General Purpose Models

Llama 2 Family

Llama 2 7B - Fast general AI 

ollama pull llama2:7b
# Size: 4.1GB | RAM: 8GB | Speed: ⭐⭐⭐⭐⭐

Best for: - Documentation writing - Technical explanations - Project planning - Quick Q&A

Llama 2 13B - Better reasoning 

ollama pull llama2:13b
# Size: 7.6GB | RAM: 16GB | Speed: ⭐⭐⭐⭐

Best for: - Complex explanations - Technical writing - Problem analysis - Requirement gathering

Llama 2 70B - Professional grade 

ollama pull llama2:70b
# Size: 40GB | RAM: 64GB+ | Speed: ⭐⭐

Best for: - Enterprise documentation - Complex technical discussions - Strategic planning - Research and analysis

Mistral Family

Mistral 7B - Efficient and capable 

ollama pull mistral:7b
# Size: 4.4GB | RAM: 8GB | Speed: ⭐⭐⭐⭐⭐

Strengths: - Excellent efficiency - Strong reasoning - Good code understanding - Multilingual support

Best for: - Technical documentation - Code reviews - API design discussions - Debugging assistance

Mixtral 8x7B - Mixture of experts 

ollama pull mixtral:8x7b
# Size: 26GB | RAM: 32GB+ | Speed: ⭐⭐⭐

Strengths: - Advanced reasoning - Multi-task capability - Excellent quality - Versatile applications

Best for: - Complex problem solving - Architecture discussions - Research and analysis - Strategic planning

Hardware-Based Recommendations

Low-End Systems (8GB RAM)

# Primary choices
ollama pull codellama:7b        # Code tasks
ollama pull mistral:7b          # General tasks
ollama pull llama2:7b           # Documentation

# Alternative options
ollama pull deepseek-coder:6.7b # Advanced code analysis
ollama pull starcoder:7b        # Code completion

Mid-Range Systems (16GB RAM)

# Recommended setup
ollama pull codellama:13b       # Primary code model
ollama pull llama2:13b          # General purpose
ollama pull mistral:7b          # Fast tasks

# Specialized additions
ollama pull codellama:7b-python # Python work
ollama pull codellama:7b-instruct # Structured tasks

High-End Systems (32GB+ RAM)

# Professional setup
ollama pull codellama:34b       # Advanced code work
ollama pull mixtral:8x7b        # Complex reasoning
ollama pull deepseek-coder:33b  # Code analysis

# Keep smaller models for speed
ollama pull codellama:7b        # Quick tasks
ollama pull mistral:7b          # Fast responses

Enterprise Systems (64GB+ RAM)

# Full capability setup
ollama pull llama2:70b          # Best general model
ollama pull codellama:34b       # Best code model
ollama pull mixtral:8x7b        # Advanced reasoning
ollama pull deepseek-coder:33b  # Code analysis

# Specialized models
ollama pull codellama:13b-instruct
ollama pull codellama:7b-python

Model Performance Comparison

Inference Speed (Tokens/Second)

Model 7B Models 13B Models 34B+ Models
CPU (16-core) 20-30 tok/s 10-15 tok/s 3-5 tok/s
RTX 3080 60-80 tok/s 30-40 tok/s 10-15 tok/s
RTX 4090 100-120 tok/s 50-60 tok/s 20-25 tok/s

Quality vs Speed Trade-offs

# For interactive development (speed priority)
ollama run codellama:7b

# For thorough code reviews (quality priority)
ollama run codellama:34b

# For balanced workflows (quality + reasonable speed)
ollama run codellama:13b

Use Case-Specific Recommendations

Web Development

# Frontend (JavaScript/TypeScript)
ollama pull codellama:13b       # React, Vue, Angular
ollama pull mistral:7b          # Documentation

# Backend (Node.js, Python, Go)
ollama pull codellama:13b       # API development
ollama pull deepseek-coder:6.7b # Performance optimization

Data Science & ML

# Python-heavy workflows
ollama pull codellama:7b-python # Pandas, NumPy, Scikit-learn
ollama pull llama2:13b          # Explanations and documentation
ollama pull codellama:13b       # Complex algorithms

DevOps & Infrastructure

# Infrastructure as Code
ollama pull codellama:13b       # Terraform, Ansible
ollama pull mistral:7b          # Documentation
ollama pull llama2:13b          # Architecture discussions

Mobile Development

# iOS/Android
ollama pull codellama:13b       # Swift, Kotlin, Dart
ollama pull mistral:7b          # Quick questions

Game Development

# Unity, Unreal, Godot
ollama pull codellama:13b       # C#, C++, GDScript
ollama pull llama2:13b          # Game design discussions

Model Update Strategy

Regular Updates

# Check for model updates monthly
ollama list
ollama pull codellama:7b    # Update main models
ollama pull mistral:7b

# Clean up old versions
ollama rm old-model-version

Version Management

# Pin specific versions for consistency
ollama pull codellama:7b-20231201

# Create custom model variants
ollama create my-coding-assistant -f Modelfile

Performance Monitoring

Track Model Performance

# Monitor resource usage
htop                        # CPU and memory
nvidia-smi                  # GPU usage
ollama show model-name      # Model info

Benchmark Different Models

import time
import requests

def benchmark_model(model, prompt):
    start_time = time.time()
    response = requests.post('http://localhost:11434/api/generate', 
                           json={'model': model, 'prompt': prompt})
    end_time = time.time()

    return {
        'model': model,
        'response_time': end_time - start_time,
        'response_length': len(response.json().get('response', ''))
    }

# Test different models
models = ['codellama:7b', 'codellama:13b', 'mistral:7b']
prompt = "Write a Python function to sort a list"

for model in models:
    result = benchmark_model(model, prompt)
    print(f"{result['model']}: {result['response_time']:.2f}s")

Troubleshooting Model Issues

Model Won't Load

# Check available resources
free -h                     # Memory
df -h                       # Disk space
nvidia-smi                  # GPU memory

# Try smaller model
ollama pull codellama:7b    # Instead of 34b

Poor Performance

# Check GPU utilization
nvidia-smi

# Adjust parameters
ollama run model --num-gpu 1     # Force GPU
ollama run model --num-gpu 0     # Force CPU
ollama run model --num-ctx 2048  # Reduce context

Model Comparison

# A/B test different models
echo "def fibonacci(n):" | ollama run codellama:7b
echo "def fibonacci(n):" | ollama run deepseek-coder:6.7b
echo "def fibonacci(n):" | ollama run codellama:13b

Next Steps