Unexpected weekly usage increase

On Aug. 19–20, while running a Codex workload, my weekly usage increased roughly 35–40 percentage points in under four hours, reaching approximately 65%. I’ve run comparable workloads many times before and would normally have expected the run to consume closer to 10 percentage points. The increase was significant enough that I stopped the runs before they finished.

I was using Sign in with ChatGPT. I read Tibo’s post on X today (https://x.com/thsottiaux/status/2090675027670978569) mentioning sub2api as a possible early finding in some of the cases OpenAI reviewed. I had never heard of sub2api before that post and have never used it. The explanation Tibo described does not align with my experience.

I wanted to add this here as another data point for the team. Based on the usage data and telemetry available to me, this run was a clear outlier compared with my previous comparable runs.

Appreciate the team looking into it. I’m happy to provide timestamps or sanitized telemetry if useful.

Start with the reported Aug. 19–20 workload and compare its roughly 35–40-point usage increase with the author's comparable runs. Review the timestamps and sanitized telemetry if provided; the issue is complete when the outlier has a reproducible cause or the evidence needed for further investigation is clearly documented.