Understand the quota clocks before they bite
Rolling, weekly, promo windows
Dual limits: 5-hour rolling + weekly reset. The +50% boost ran 2026-05-13 to 2026-09-13, ended (@ClaudeDevs 2026-08-29 said 2026-09-14). From 2026-09-14 standard limits sit 25% above pre-promotion. Not the same as cost-tuning.
The two clocks
A 5-hour rolling window plus a fixed weekly reset. The help centre, fetched 2026-09-20, reads in the past tense: It ran from May 13 through September 13, 2026 — the temporary +50% weekly boost that began 2026-05-13 closed that day. The same page says starting September 14, 2026 weekly limits are 25% higher than before. The @ClaudeDevs post of 2026-08-29 named 2026-09-14, one day apart; keep both dates with their source. Dashboard usage and key stats show where you stand.
Promo window +50% (ended 09-13)
Weekly headroom
Help Center fetched 2026-09-20 reads past tense: the extra 50% weekly quota ran 2026-05-13 to 2026-09-13; from 2026-09-14 standard weekly limits are 25% above pre-promotion. The @ClaudeDevs post of 2026-08-29 named 2026-09-14, one day later.
Observe in dashboard
Track remaining and reset times per key to plan heavy sessions.
Rolling 5h + fixed weekly
Two independent limits create the common confusion. Rolling window refills gradually; the weekly one is a hard boundary for most users.
Hedging in practice
Keep part of your traffic spread across Grok, GPT tiers and Kimi, so a single provider's reset never stops all work. A self-built Grok Build with a local model can absorb baseline tasks.
QCode makes hedging trivial
One key reaches Claude, GPT tiers and Chinese models. Split traffic when a window tightens. A self-managed local endpoint can carry privacy-sensitive baseline work.
Practical checklist
- Check both rolling and weekly numbers before long agent runs.
- Keep a fallback model ready in your routing rules.
- The +50% promo ran 2026-05-13 to 2026-09-13 and ended (the @ClaudeDevs post named 2026-09-14). Weekly limits sit 25% above pre-promotion, our math ~17% under promo headroom. Recheck the dashboard; don't plan capacity on promo headroom.
Common pitfalls
Treating the 5-hour window and weekly limit as one clock; treating the +50% as standard quota (ended 2026-09-13 per the help centre, 2026-09-14 per @ClaudeDevs; limits run 25% above pre-promotion, our math ~17% under promo headroom); no cross-model fallback.
QCode makes hedging trivial
One key reaches Claude, GPT tiers and Chinese models. Split traffic when a window tightens. A self-managed local endpoint can carry privacy-sensitive baseline work.
Quota FAQ
How do I see my current window?
QCode dashboard and the provider usage endpoints show remaining and reset timing per key.
Does the promo affect all users?
Help Center eligibility: Pro, Max, Team, and legacy seat-based Enterprise. Free and consumption-based Enterprise are excluded. 5-hour limits are unchanged. Fetched 2026-09-20 the page reads past tense: +50% ran 2026-05-13 to 2026-09-13, and starting 2026-09-14 weekly limits are 25% higher than before; @ClaudeDevs on 2026-08-29 named 2026-09-14. Our own math: about 17% under promo headroom. Plans and billing do not change with this promo.
Is this the same as cost optimization?
No. Cost opt teaches spend tricks. This page is about hard limit timing and provider switching to keep velocity.
Can I mix models to avoid resets?
Yes. One QCode key reaches multiple families; route around the tight window.