GPT-5-mini is 0x or 1x premium request?

[eduardosantos1989]

Describe the bug

I tried to run a simple copilot prompt to test gpt-5-mini and in the end it turned out that it counted for me as 1 Premium request however it says it should count 0 so what is the reality?

copilot --allow-all-tools --resume --model gpt-5-mini -p "please curl ifconfig.me"
Running curl to ifconfig.me to fetch the host's external IP address.

✔ $ curl -s ifconfig.me || curl -s https://ifconfig.me/ip
    └ 2 lines...

External IP returned by ifconfig.me: 60.110.86.153


Total usage est:       1 Premium request
Total duration (API):  8s
Total duration (wall): 14s
Total code changes:    0 lines added, 0 lines removed
Usage by model:
    gpt-5-mini-2025-08-07 81.9k input, 512 output, 57.3k cache read (Est. 1 Premium request)

Affected version

0.0.366 Commit: 0c18d21

Steps to reproduce the behavior

run a simple prompt against gpt-5-mini

Expected behavior

Total usage est: 0 Premium request

Additional context

No response

[jbeattie5768]

I saw this today too (first time I've used the CLI), but I don't believe it's actually billing you 1x for gpt-5-mini, assuming you have a paid plan.

Checking my Copilot usage in my GitHub settings:

• % Usage (coarse scale): https://github.com/settings/copilot/features
• Usage Count (slow to update): https://github.com/settings/billing/usage
  • You can download the Usage Count as a CSV as well, but it gives no more info than given in the table

I do not see it increase for gpt-5-mini. I do see it rise when it uses the default claude-sonnet-4.
I'd say the GH usage view reflects the billing system directly, not the CLI’s estimate.

If you are using a "Copilot free", then the multiplier is 1x, according to this page at least: https://docs.github.com/en/copilot/concepts/billing/copilot-requests#model-multipliers. This page also matches what I see as the paid-plan multipliers in VSCode.

The CLI is wrong in its estimation for gpt-5-mini and paid plans, but I believe the actual billing is not affected.

Checking:

gpt-4.1:            0x    prints "0 Premium requests" <-- correct output
gpt-5-mini:         0x    prints "1 Premium request" 
gpt-5.1-codex-mini: 0.33x prints "1 Premium request"
claude-haiku-4.5:   0.33x prints "1 Premium request"
claude-opus-4.5:    1x    prints "1 Premium request"  <- webpage claims 3x after 5th Dec

I see the billing increase by 1 request for the 0.33x multiplier models, but I suspect that is just the coarseness of the count and not the actual billing (?).

So there are definite discrepancies in the CLI estimates for paid plans, but I don't think this is reflected in the actual billing. Be good to get it confirmed, though!

[andyfeller]

Thank you both for raising and collaborating on this issue, @eduardosantos1989 @jbeattie5768! ❤️

I want to confirm that you aren't actually being billed for the gpt-5.1-mini usage; this is simply a bug due to how the requested model was being reported back. There is a fix up for review as we speak! 🙌