Updated April 2026
GitHub Copilot Extensions, Agent Mode, and Premium Requests:
What Costs Extra (2026)
The shift to "premium requests" is the most confusing change in Copilot pricing. Here is exactly what counts against your monthly quota, what does not, and how to estimate your usage before upgrading.
What Uses Premium Requests
Consumes premium requests
- •
Copilot Chat messages
1 PRU per message
- •
Agent mode turns
1 PRU per step, complex tasks 5-20 PRUs
- •
Code review requests
1 PRU per review
- •
Premium model selection
GPT-4o, Claude Sonnet, Gemini - 1 PRU per message
- •
GitHub.com Chat
Enterprise only - 1 PRU per message
- •
Copilot Extensions interactions
1 PRU per Chat message to extension
Does NOT consume premium requests
- •
Inline code completions
Tab suggestions are always free (unlimited on paid plans, 2,000/mo on Free)
- •
Default model (GPT-4o-mini)
The base model on Free tier does not cost PRUs
- •
Navigating Copilot interface
Viewing suggestions, settings, etc.
- •
Rejected suggestions
Viewing and dismissing completions does not count
Premium Request Allocation by Plan
| Plan | PRUs/month | Overage cost | Approx. daily Chat capacity |
|---|---|---|---|
| Free | 50 | $0.04/request | ~2-3 messages/day (20 working days) |
| Pro | 300 | $0.04/request | ~15 messages/day |
| Pro+ | 1,500 | $0.04/request | ~75 messages/day |
| Business | 300/seat | Org billing | ~15 messages/day per developer |
| Enterprise | 1,000/seat | Org billing | ~50 messages/day per developer |
Estimate Your Monthly Premium Request Usage
Enter your typical daily usage to see which plan fits your usage pattern.
Daily usage inputs
Assumes ~8 steps per session on average
Monthly PRU breakdown
Estimates assume 20 working days/month. Agent mode step count varies widely by task complexity. Track your actual usage in GitHub Settings > Copilot to verify.
Agent Mode and PRU Consumption
What agent mode does
Agent mode allows Copilot to autonomously determine which files to edit, run terminal commands, and iterate on errors. It is available in VS Code and JetBrains as of March 2026.
Each step the agent takes consumes one premium request. A complex task - refactoring a module, setting up a new feature, debugging a failing test - can take 5-20 steps.