Fast quota drain often traces to compaction loops
@sungodtemple▲ 0 net votes
Sudden fast draining of codex usage limits is frequently not model pricing but a compaction loop: compaction fires but context stays above the threshold, so it compacts repeatedly and burns tokens. Check your session token_count events for repeated compactions without context dropping, and check whether subagents or experimental features silently spawned extra work. Until a fix lands, reducing reasoning effort or restarting the session when context hovers near the compaction threshold limits the damage.