Reduce GitHub Copilot token usage in Visual Studio Code
Fokko at Work explains how to cut GitHub Copilot token usage in Visual Studio Code after the shift to usage-based billing, sharing five practical tweaks that reduce AI credit consumption without sacrificing output quality.
Overview
Halfway through 2026, GitHub Copilot introduced usage-based billing, meaning token consumption (AI credits) now matters more for many users. This video walks through five concrete ways to reduce token usage in VS Code while keeping results useful.
Topics covered
Use tool output compression
- Reduce the amount of tool output Copilot has to process.
- Aim to keep only the relevant parts of logs/outputs when feeding context back into Copilot.
Optimize model choice
- Pick a model that matches the task instead of defaulting to the most expensive option.
- Use GitHub Copilot's documentation to compare models and understand trade-offs.
Links discussed:
- Auto model selection: https://docs.github.com/en/copilot/concepts/models/auto-model-selection
- AI model comparison: https://docs.github.com/en/copilot/reference/ai-models/model-comparison
Optimize for cache usage
- Structure prompts and repeated work so Copilot can benefit from caching.
Link discussed:
- Blog post on caching: https://www.linkedin.com/pulse/hidden-life-tokens-github-copilot-fokko-veegens-yf22e/
Be concise
- Keep prompts short and specific to reduce unnecessary token spend.
- Avoid sending large, unfocused context when a smaller excerpt will do.
Video chapters
- 00:00 Intro
- 00:24 Stay updated
- 02:02 Use tool output compression
- 04:47 Optimize model choice
- 12:27 Optimize for cache usage
- 19:08 Be concise
- 22:08 Outro