In this article, Rob Bos delves into the recent changes around billing for Generative AI in GitHub Copilot, emphasizing the new Premium Requests pricing model and its implications for developers.

Introduction

GitHub Copilot users will now be directly responsible for costs linked to Generative AI usage, especially for requests made to non-default models. This is a shift from the previous, largely opaque, model where usage felt ‘free.’ The change is designed to encourage accountability and a better understanding of the resource consumption and its real monetary cost.

What Are Copilot Premium Requests?

Starting June 18th, Premium Requests take effect for Copilot. These are charges for using models other than the base/default model (such as GPT-4o, GPT-4.1). Requests to advanced models (e.g., GPT-4.5) can be up to 50x costlier than the base, whereas some more efficient models (Gemini 2.0 Flash) have a reduced cost multiplier.

Actions That Consume Premium Requests

  • Chatting with any non-default model: 1 premium request per “turn.”
  • Steps performed by the Coding Agent: 1 per step, possibly multiple per conversation. Max requests setting is anticipated.
  • Code review in pull requests: 1 per request.
  • Agent Mode in editors: 1 per user-initiated request.
  • Copilot Coding Agent: may use multiple requests as needed.
  • Copilot Spaces and Copilot Extensions usage: likewise consume premium requests, details somewhat unclear.

Pricing and Usage Across Plans

  • Free Plan: Even the base model’s chat costs premium requests (50/month included).
  • Paid Plans:
    • Copilot Pro: 300/month, unlimited with base model.
    • Copilot Pro+: 1500/month
    • Copilot Business: 300 per user/month
    • Copilot Enterprise: 1000 per user/month Additional requests are billed at .04 each, with multipliers applying according to model selection. For example, GPT-4.5 at 50x multiplier = $2/request.

Budget Controls and Usage Tracking

End-users and organizations can set monthly budgets; when this is hit, premium requests pause until the next cycle. Editors like VS Code and JetBrains now show model cost multipliers and usage info directly in the UI, allowing users to choose the most cost-effective model for their needs.

Analyzing and Managing Usage

Maintenance and monitoring are supported via CSV exports from the Copilot dashboard. Rob provides a GitHub-hosted, open-source analyzer SPA (using Spark and Copilot Coding Agent) to make sense of usage data, crucial for organizations to audit and optimize Copilot spending.

Conclusion

Rob Bos emphasizes the need for proactive management with these changes: users should monitor their premium request usage, set budgets, and leverage in-editor usage cues to prevent unexpected charges. As this pay-per-use paradigm enforces more responsible usage, tools and transparency enhancements play a critical role in user experience.

For more details and tooling, see GitHub Copilot Premium Requests Usage Analyzer.

This post appeared first on Rob Bos’ Blog. Read the entire article here