Single user prompt of haiku-4.5 is counted as 33 “Premium requests” instead of 1

[ming86]

Describe the bug

After updating to GitHub Copilot CLI v0.0.344 (commit c87f222), a single interactive prompt is reported as consuming ~33 Premium requests (x 0.33), instead of 1 Premium request (× 0.33x). The same workflow on the previous version worked as expected yesterday.

I did not tried other models.

Affected version

0.0.344 Commit: c87f222

Steps to reproduce the behavior

Launch Copilot CLI interactively (e.g., gh copilot or your usual entrypoint).

Submit one prompt, for example:

Please mask the token that logged on the debug log.

Let Copilot complete the task and show the session usage summary, and view the Copilot usage.

Expected behavior

For a single user prompt haiku-4.5, should consumes 1 Premium request × 0.33x model multiplier, not dozens.

Additional context

Session Output (redacted)

Welcome to GitHub Copilot CLI
 Version 0.0.344 · Commit c87f222
 ...
 ● Logged in with gh as user: <username>
 ● Connected to GitHub MCP Server
 > Please mask the token that logged on the debug log.
 ...
 ● Total usage est:       31 Premium requests
   Total duration (API):  1m 36.1s
   Total duration (wall): 27m 22.4s
   Total code changes:    112 lines added, 4 lines removed
   Usage by model:
       unknown              0 input, 1.0m output, 0 cache read, 0 cache write (Est. 31 Premium requests)
 ...
 claude-haiku-4.5 (0.33x)

Note: Yesterday’s version did not overcount.

[Andrej730]

#351

[nathfavour]

same issue here. quite unheard of.

[RyanHecht]

Hey folks -- going to continue tracking this in #351 -- thanks for the reports!