Codex Rate Limits Discussion Thread

Hi OpenAI Team and Community, Here is my experience regarding an issue of fast credit consumption by Codex (usage in India). It consumed credits worth Rs 1,700 twice, in less than 5 hours for each credit top-up. I am on the Plus plan and use the Codex desktop tool, working on a debugging task. I don’t feed files in the prompt, as Codex has access to the development folders on disk. Codex operates in agent mode, as it has been given approvals except for unsafe tasks. The Codex workspace/sandbox contains a few documents, such as project-status.md and issue-tracker.md, which are read and written during every mini-session of up to 10–12 minutes. My current work involves debugging code that was ported from C++ to C#. Codex runs build commands on its own. It doesn’t use the cloud and relies only on the local disk. It accesses the internet only when I ask it to check something online. Comparing this work with last month’s work: One distinctly different aspect is that last month as well, I worked on coding tasks across three sub-modules involving C++ and Python. This was not a porting effort but involved new development work, debugging, build and testing. Approvals were set to the default mode, i.e., ask for every command execution. During that period, I was also on the Plus plan, but I never experienced reaching usage limits, needing to add credits, or rapid credit consumption. In the last two days, it consumed credits worth Rs 1,700 twice, in less than 5 hours for each credit top-up. That’s unbelievable as the monthly subscription of Plus plan cost Rs 1,999 for Indian users. It’s like hitting usage limit in few hours of taking subscription. All comments and suggestions are welcome. Thanks. Best Regards, -Darshan Maharana