Repository navigation
Large unexpected spike on default (sonnet 4.5) model usage on #1470
Description
Describe the bug
I've noticed a huge spike on my sonnet 4.5 usage today whilst using auto pilot on claude-opus-4.6
This seems to have triggered something sending off a large number of requests on sonnet 4.5 - is there any known issues on 410 which may have triggered this?
My session was also leveraging the Copilot SDK iterating on some prompts on gpt-5-mini so it is possible there was an issue with defaults being picked up there but i don't have an easy mechanism to diagnose which may have triggered at the moment.
Affected version
v0.0.410
Steps to reproduce the behavior
Seemed to occur whilst running opus 4.6 on Auto pilot
Expected behavior
Standard opus request consumption - no Sonnet 4.5 (And not in the region of 500+ requests)
Additional context
No response
Activity
patniko commentedon Oct 26, 2026
Thanks for the report, and sorry this went unanswered for so long. We weren't able to investigate without session details, and a lot has changed since v0.0.410: Copilot has moved to usage-based billing with AI credits,
/usagenow breaks down consumption per model (v1.0.85), you can see and configure the models subagents use with/subagents(v1.0.62), and autopilot looping issues such as #1532 have been fixed. Usage on a model you didn't select can come from subagents, or from SDK sessions created without an explicitmodel, which use your default model. If you still see unexpected usage on the latest version (v1.0.103), please open a new issue with your CLI and SDK versions, the/usageoutput, how each session was started (including the SDK session config), and debug logs (--log-level all --log-dir <dir>).Triaged as part of a Copilot SDK backlog review, with help from GitHub Copilot.