01 · The 5-hour window: same idea, one difference
On both, the window opens with your first message and closes 5 hours later — it is not tied to the clock, and everyone's start time is different. Claude Code drains it on a rolling basis (usage from ~5 hours ago frees up gradually); Codex on Plus is a hard bucket that empties at the 5-hour mark.
The one structural difference: Codex Pro ($100 and $200) dropped the 5-hour gate in August 2026, leaving only the weekly allowance. Every Claude plan, Max included, keeps the 5-hour session.
02 · Weekly reset: rolling vs fixed
This is the difference people trip over. Codex's weekly window rolls from your first request after the previous reset — use it on Tuesday and it resets the following Tuesday, use it on Friday and it moves. Your billing date is irrelevant.
Claude Code's weekly allowance resets at a fixed time each week assigned to your account, visible under Settings → Usage. It does not move, no matter when you start using it. If you plan work around a known reset time, Claude is predictable; if you want a fresh week whenever you come back from a break, Codex's rolling window is friendlier.
03 · Plans side by side
Codex: Plus ($20) is gated by both windows. Pro $100 is roughly 5× Plus, Pro $200 roughly 20×, and both drop the 5-hour gate. Business / Enterprise seats default to roughly Plus level.
Claude Code: Pro ($20) gives at least 5× the free tier per session. Max 5x ($100) is about 5× Pro, Max 20x ($200) about 20× Pro. Both Max tiers meter Opus on its own 5-hour and weekly pools, so heavy Opus work does not eat your Sonnet allowance — there is no Codex equivalent of a per-model pool.
Neither publishes absolute token counts. Both moved to token-based metering in 2026 (Codex in April; Claude has always metered input + output tokens), so "messages per day" figures you find elsewhere are outdated.
04 · How often each one actually resets everyone
This is where the two companies differ most, and it is the part no pricing page tells you. OpenAI's Codex lead announces fleet-wide resets on X — after outages, at user milestones, at model launches, and as compensation for metering bugs. Anthropic does the same on @ClaudeDevs, but far less often, usually around a model launch or an incident.
The numbers below come straight from our records and update as we log new resets. In practice, a Codex Plus subscriber gets meaningfully more allowance than the plan states because of these resets; a Claude subscriber should plan on the stated allowance.
05 · Banked resets
Only Codex has them. OpenAI sometimes issues a reset credit you redeem yourself from the profile menu instead of applying it immediately; they stack, expire 30 days after issue, and redeeming one refreshes both windows. Anthropic resets are always applied directly — there is nothing to save or time.
06 · Checking where you stand
Codex: /status in the CLI shows both bars; /usage shows banked resets and their expiry; on desktop, avatar → Usage. Claude Code: /usage shows both windows (and Opus separately on Max); when you are limited, the message itself states the reset time.
07 · So which one?
If you work in bursts and take breaks, Codex Plus plus its frequent free resets is the better deal per dollar. If you run long agent sessions every day and want a predictable weekly reset — or you want Opus without touching your main allowance — Claude Max is built for that. At $200, Codex Pro's missing 5-hour gate and Claude Max 20x's separate Opus pool are the two things to weigh.
Either way, the top of each provider page here tells you the last time that company reset everyone, and the "since last reset" figure tells you how the current wait compares to history.
// AND THE OTHER FIVE
Grok uses a single weekly compute pool with no 5-hour window. Gemini CLI mirrors the 5-hour + weekly shape but counts model requests. Kimi Code and GLM Coding Plan are the cheapest with the same two-layer shape (GLM counts prompts). Cursor is the odd one out: a monthly dollar pool that resets on your billing date.