After Switching from PRU to Usage-based, Copilot is have lot of bug

[ourcaldo]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

Copilot Enterprise

Body

Hi all,

I'm an enterprise admin, and one of our members is using Copilot after we switched from PRU to usage-based billing.
My team has been using Copilot as usual. This particular member only asked 2 questions about something, and then the usage was completely full.

I tried to understand the issue, and he gave me the history of his session where each conversation took around 15–20 credits at most. But here's the issue: when I tried to download the usage report, I saw the usage was literally more than 1,000 credits on his very first request.

What exactly happened here? We need a clarification from you, GitHub!

Thanks,

ScreenRecorderProject2.mp4

Hello,

We have provided more details and a FAQ about usage-based billing in this discussion, which may be updated if new information becomes available.

To keep conversations all in one place and to ensure you receive timely updates, we’ll be closing and locking this post and recommend you comment on or subscribe to the discussion linked above. Thank you!

@ourcaldo - Apologies for pre-emptively closing this post. That was my mistake.

We've looked into what the usage report — thank you for the context. Validating what you shared: The discrepancy between what your team member sees in their chat history (15–20 credits per conversation) and what the CSV reports (1,000+ credits on a single request) suggests that additional processing is occurring behind the scenes that isn't visible in the user's session, potentially related to tool context loading and agent sub-requests.

We're aware of community reports identifying this pattern and the engineering team is investigating. We don't have an ETA for any updates, but anything we do share will be on the discussion linked in the previous comment. Again, apologies for the additional confusion and frustration.

Next steps: This forum is not a support queue. If you'd like to formally report this as a bug with your CSV evidence, please file at:
https://github.com/github/app/issues. Your usage CSV showing the disparity is excellent supporting evidence.

If helpful, here are other suggestions:

1. Reduce overhead: Have the affected user disable unused built-in tools and extensions in VS Code (Settings → Extensions). This has been shown to reduce per-request token baseline significantly.
2. Switch from agent mode to standard chat — agent mode triggers multiple internal sub-requests that accumulate cost.
3. Use lighter models where possible — Claude Sonnet 4.6 is among the most expensive models available.