Codex usage limits seem to be draining unusually fast since 10 May 2026, even with light GPT-5.3 Codex use, while the usage-tracking UI also appears buggy or missing

I agree with you. And I’d add one more observation from my own experience. Codex doesn’t live in one dimension — it lives in three. And the user pays for all three simultaneously without a clear breakdown. What is officially confirmed by OpenAI: AGENTS.md is read at the start of every session. Source: OpenAI documentation. Larger codebases and longer sessions consume significantly more per message. Source: OpenAI Help Center. Large system files reduce how far a limit lasts. Source: OpenAI documentation. What OpenAI has never published: Exactly which files get loaded at session startup beyond AGENTS.md. Whether git operations like commit, diff, or push consume tokens. Whether reading screenshots or images costs tokens and how much. How many tokens equal one percent of the 5-hour window. And that silence is the problem. Not the fact that tokens are consumed — that’s expected. The problem is that there is no public breakdown. No table. No estimate before a task starts. No receipt after it ends. The user has no way to plan, predict, or verify what they’re actually paying for.