'generate copilot summary' consumes too many premium requests?

[chenluyao-mmind]

I am using GitHub Copilot Enterprise in Visual Studio Code and noticed that the code summarization feature consumes an unexpectedly high number of premium requests.

A single "Generate copilot summary" action typically uses around 30 premium requests, which seems excessive for one operation. And no option to control model or context size.

[benmcmorran]

I haven't been able to reproduce this issue, but it's certainly not the expected behavior. The feature uses VS Code's Language Model API to send a single request to the LLM, using the first available model provided by VS Code. One possible explanation is that model selection is defaulting to Claude Opus 4.6 (fast mode), which currently has a 30x multiplier.

I opened #14349 with a potential fix to prefer GPT-5-mini for this feature, which has a 0x multiplier on paid plans.