GitHub Senior Developer Advocate Andrea Griffiths breaks down the differences between GitHub Copilot and raw API access to large language models, helping developers understand what they're paying for with each option.
Copilot's integrated experience
Copilot offers a seamless, context-aware coding assistant that understands your project's codebase, dependencies, and coding style. It provides inline suggestions, chat-based help, and integration with GitHub's ecosystem. The subscription covers the cost of model inference, prompt engineering, and continuous improvement based on user feedback.
Raw API access
Using an LLM API directly gives you full control over the model, prompts, and integration. You pay per token and must handle context management, prompt crafting, and output filtering yourself. This approach is more flexible but requires significant engineering effort to achieve similar results.
Key trade-offs
- Context awareness: Copilot automatically includes relevant code context; raw API requires manual context injection.
- Ease of use: Copilot works out of the box in supported IDEs; raw API demands custom integration.
- Cost structure: Copilot has a flat monthly fee; raw API costs scale with usage.
- Customization: Raw API allows fine-tuning and custom models; Copilot is limited to GitHub's models.
- Privacy: Both offer enterprise options with data not used for training.
What you're paying for
With Copilot, you pay for convenience, integration, and reduced cognitive load. With raw API, you pay for flexibility and control, but bear the cost of building and maintaining the assistant layer. Griffiths suggests that for most developers, Copilot's value proposition lies in saving time and effort, while raw API is better suited for specialized use cases or when you need to use a model not supported by Copilot.