Using Thox.ai Coder Models

Complete guide to the 7B, 14B, and 32B Thox Coder models and how to use them effectively.

Back to Articles

New: Thox.ai Cluster Code

For software engineering teams, Cluster Code is our elite distributed model with GPT-4o competitive performance. Runs on 4-8 clustered devices, supports 92 programming languages, 128K context for repository-level analysis, and serves 15-30 concurrent engineers with 100-150 tokens/sec per user.

Learn About Cluster Code

Thox.ai comes with three purpose-built coding models optimized for different use cases. Each model is fine-tuned on millions of code repositories and optimized to run efficiently on your local hardware. Choose the right model based on your task complexity and performance needs.

Model Comparison

ModelContextMemorySpeedBest For
Coder 7B8K tokens~8 GB45-60 t/sReal-time completion
Coder 14B16K tokens~16 GB25-35 t/sBalanced tasks
Coder 32B32K tokens~36 GB12-18 t/sComplex reasoning

Model Details

Thox Coder 7B

7 billion parameters

Fast and efficient model optimized for quick code completions, simple refactoring, and real-time suggestions. Ideal for everyday coding tasks.

Specifications

Context Length:8,192 tokens
Memory Required:~8 GB
Generation Speed:45-60 tokens/s
First Token Latency:<100ms first token

Best For

  • Real-time code completion
  • Quick syntax suggestions
  • Simple function generation
  • Fast iteration during development

Thox Coder 14B

14 billion parameters

Balanced model offering improved reasoning and code understanding while maintaining good performance. Great for most development workflows.

Specifications

Context Length:16,384 tokens
Memory Required:~16 GB
Generation Speed:25-35 tokens/s
First Token Latency:<200ms first token

Best For

  • Complex code completion
  • Code refactoring
  • Bug detection and fixing
  • Documentation generation
  • Unit test creation

Thox Coder 32B

32 billion parameters

Our most capable model with advanced reasoning, multi-file understanding, and complex problem-solving abilities. Best for challenging tasks.

Specifications

Context Length:32,768 tokens
Memory Required:~36 GB
Generation Speed:12-18 tokens/s
First Token Latency:<400ms first token

Best For

  • Complex algorithm implementation
  • Multi-file refactoring
  • Architecture suggestions
  • Code review and analysis
  • Advanced debugging
  • Full feature implementation

Recommended Use Cases

Code Completion

Get intelligent suggestions as you type with context-aware completions.

Recommended: 7B for speed, 14B for accuracy

Refactoring

Restructure and improve existing code while maintaining functionality.

Recommended: 14B or 32B for complex refactoring

Code Generation

Generate entire functions, classes, or modules from descriptions.

Recommended: 14B for simple, 32B for complex

Code Review

Analyze code for bugs, security issues, and best practices.

Recommended: 32B for thorough analysis

Switching Models

Use the CLI or web interface to switch between models:

thox models list

View installed models

thox start -m thox-coder-14b

Start with a specific model

thox models pull thox-coder-32b

Download additional models

Tips for Best Results

Match model to task complexity

Use 7B for simple completions, 14B for everyday coding, and 32B for complex reasoning tasks.

Context window matters

For multi-file tasks, use 14B or 32B which have larger context windows to understand your full codebase.

Related Articles

CLI Command Reference

All available CLI commands

Performance Optimization

Get the best speed from your models

Need help choosing?

Our support team can help you select the best model for your specific workflow.

Run Your First ModelContact Support