Usage wiki: what counts toward your limit

One question per page, answered for Codex and for Claude from the official docs. Quoted, linked, and dated.

  1. How is the 5-hour window counted?Codex opens a fixed 5-hour window at your first message after the previous one ends. Claude runs a rolling 5-hour session. Neither is a fixed number of messages.
  2. How is the weekly limit counted?Claude's weekly limit resets at a fixed time each week assigned to your account. Codex's weekly period runs 7 days from your first request after a reset; where a normal period starts is not documented.
  3. Does thinking count toward the limit?Yes, on both. Reasoning tokens are usage; a higher reasoning effort or thinking budget uses more of the window.
  4. Do subagents count toward the limit?Yes, on both, and they cost more than a single agent: each subagent does its own model and tool work from the same allowance.
  5. Does prompt caching count toward the limit?Cached input still counts, at a lower rate, on both. Codex does not charge for cache writes; Claude does not count reused project content again. Neither publishes the exact discount inside a subscription window.
  6. Does web search count toward the limit?Claude: yes, web search and web fetch count. Codex: OpenAI does not document web search separately; tool use and retrieval are listed as things that raise usage.
  7. Does the countdown change after I use a reset card?Codex: yes. A banked reset refreshes both the 5-hour and weekly windows and moves your weekly reset date. Claude has no reset cards; an announced reset clears usage, and Anthropic does not document how the timers move.

Quotes checked against the live docs on 2026-09-15.

Answers rest on official OpenAI and Anthropic documentation, quoted and linked on each page · Not endorsed by either.