The Tibo reset era bought Plus builders a summer of long Codex sessions. That window closes tomorrow for ChatGPT Plus — not for everyone.
On August 25, 2026 (~6:46 AM), OpenAI Codex and ChatGPT lead Tibo Sottiaux posted on X that OpenAI will bring back the rolling 5-hour usage limit for Plus accounts across ChatGPT Work and Codex starting the next day. He had teased the change before, then postponed it. Pro $100 and Pro $200 subscriptions keep the 5-hour gate off for the upcoming months.
The thread drew roughly 115K views within hours. Replies on X and Hacker News split three ways: bait-and-switch anger from people who subscribed while the cap was gone, a Pro vs Plus compute argument about who deserves uninterrupted agent time, and a quieter line from builders who said the missing 5-hour limit was why they picked Codex over Claude — and now have to re-evaluate.
This is the policy chapter after July's temporary removal, the July 29 reinstatement tease, and this weekend's unrelated quota-drain bug reset. Same meter, new tier split.
TL;DR
| Question | Direct answer |
|---|---|
| Who gets the 5-hour cap back? | ChatGPT Plus on Work and Codex — effective August 26, 2026 |
| Who stays uncapped at 5 hours? | Pro $100 and Pro $200 — 5-hour limit stays disabled for upcoming months |
| Why now? | Smooth compute load + keep weekly limits generous; stop casual Plus users from accidentally burning a full week |
| Was this surprise? | Tibo says he mentioned it before and postponed — not a stealth nerf, but feels like one if you joined during the pause |
| Codex vs Claude? | Plus Codex will look like dual 5h + weekly again — the edge some builders cited over Claude Code |
| End of Tibo resets? | No — resets fix bugs and milestones; policy caps are separate. Expect fewer "free refills" to paper over a hot 5-hour bucket |
What Tibo posted — full quote
Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex. I had mentioned this a while ago, but then postponed it.
This is necessary as (a) the 5h limit allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience.
We are for the upcoming months keeping the 5h limit not enabled for Pro $100 and Pro $200 subscriptions.
Source: Tibo Sottiaux on X, August 25, 2026.
Read that carefully. This is not "we ran out of GPUs." It is load smoothing so weekly Plus allowances can stay high, plus UX guardrails for users who treat a week of quota like a single afternoon sprint and then wonder why the bar is empty on Wednesday.
The timeline — why people call it bait-and-switch
| Date | What happened |
|---|---|
| July 12, 2026 | Tibo temporarily removes the 5-hour cap on Plus, Business, and Pro during Sol launch week — full July guide |
| July 29, 2026 | Tibo says the 5-hour window returns July 30 after Sol efficiency fixes — 18% longer usage + cap return |
| August 9–24, 2026 | Manual Tibo resets for milestones, proxy drama, and product drains — not the same as removing a published cap |
| August 25, 2026 | Plus-only reinstatement announced; Pro $100/$200 explicitly exempt for months |
If you subscribed to ChatGPT Plus in July or August because Codex felt unlimited inside the week, Tibo's own words confirm the 5-hour gate was always coming back for Plus — just on a delayed schedule. That is legally consistent with "temporary removal." It still reads as a subscription decision under stale terms if your mental model was "weekly only forever on Plus."
Pro subscribers get a different story: Tibo is rewarding the tier that already pays for heavier compute by leaving the 5-hour interrupt off for now. That matches OpenAI's $80 usage reset on Pro pattern — retention gifts and tier splits, not one flat policy for every paid seat.
What people are arguing about
"I chose Codex because Claude has a 5-hour wall"
That showed up in HN and X replies for a reason. Anthropic's Claude Code still runs 5-hour and weekly buckets on Pro — explainx.ai mapped that in the Claude Code pricing guide. For six weeks, Codex on Plus let you run until the weekly cap, which mattered for long loop engineering sessions and multi-hour Sol agent runs.
After August 26, Plus Codex matches Claude's dual-limit shape again. The comparison in Codex vs Claude Code shifts back toward "Codex for execution and computer use, Claude for planning" — not "Codex because the cap is gone."
Pro vs Plus — who gets uninterrupted time?
Tibo's split is blunt: casual and new Plus users get a guardrail; Pro $100/$200 keep marathon mode. If your workflow is "start a goal Friday night and wake up to a PR," Plus is the wrong SKU starting tomorrow. The cheapest frontier coding subscription math still favors Plus on cost per task — but not on session length.
End of the Tibo reset meme?
No — resets and caps are different levers. Tibo still ships full refills when product bugs drain meters (August 24 reset) or when milestones hit (20M banked credit). A restored 5-hour window means you can hit a local ceiling even with weekly headroom left — the complaint that triggered July's temporary removal in the first place.
What the 5-hour limit actually does on Plus
After August 26, Plus on ChatGPT Work and Codex should again enforce two governors:
- Rolling 5-hour window — local messages, cloud tasks, and code reviews share one bucket that refills continuously across five hours (see ChatGPT Work vs Codex for how Work and CLI count against the same pool).
- Weekly allowance — the longer-horizon cap Tibo wants to keep "generous" by throttling burst usage.
Official Codex pricing pages already list per-model ranges inside the 5-hour window (e.g. GPT-5.6 Sol local messages between roughly 15–90 on Plus depending on effort and model). The weekly number is the one casual users accidentally zero out — Tibo's point (b).
For builders, the practical shift is session planning:
- Before: one weekly bar; long sessions limited only by Sol burn rate and token-heavy side features.
- After (Plus): check both bars; a Friday night agent run can exhaust the 5-hour bucket even if Monday's weekly balance looked healthy.
Pro $100/$200 users skip the first bullet for now — still subject to weekly limits, still sharing Work + Codex quota, still vulnerable to product-side drains.
Codex vs Claude — updated decision table
| If you optimize for… | Plus Codex after Aug 26 | Claude Pro ($20) |
|---|---|---|
| Long uninterrupted sessions | 5h cap returns — plan breaks or upgrade to Pro | 5h + weekly — same wall, different model taste |
| Weekly total usage | OpenAI says weekly stays generous; 5h smooths bursts | Anthropic extended limits through promos — check current banner |
| Cost per coding task | Still strong on Sol Medium — subscription comparison | Higher intelligence per task on Opus tiers — pricing guide |
| Harness / tool breadth | Codex CLI, Work, computer use — harness guide | Claude Code CLI, IDE, subagents — commands reference |
Honest take: if the only reason you were on Plus Codex was "no 5-hour wall," re-run the comparison this week. Many serious builders already run both at ~$20 each; this change nudges marathon users toward Pro or Claude Max, not away from agents entirely.
What to do today
- Check your tier — Plus vs Pro $100/$200. Only Plus gets the cap tomorrow.
- Screenshot your meters before August 26 if you are debugging usage — baseline helps if the UI jumps.
- Split long goals into sub-5-hour blocks, or move heavy 1M-context experiments to Pro if you have it.
- Use Sign in With ChatGPT on official clients — shared proxies still trigger fraud flags unrelated to this policy.
- Budget tokens deliberately — token budget planning matters more when a local 5-hour ceiling returns.
If you are deciding between upgrading to Pro or switching harnesses, treat Tibo's post as pricing signal, not emergency: Pro keeps the cap off for months, not permanently.
FAQ
The frontmatter FAQ schema covers the searchable questions; the short version:
- Plus: 5-hour limit returns August 26.
- Pro $100 / Pro $200: 5-hour limit stays off for upcoming months.
- Why: compute smoothing + protect casual users from burning a week accidentally.
- Claude comparison: Plus Codex loses a real differentiator; dual limits look similar again.
Related on explainx.ai
- Why Codex quota drained fast — Tibo reset Aug 24
- ChatGPT drops 5-hour limit — Tibo Jul 12
- Sol ~18% longer + 5-hour return tease — Jul 29
- Cheapest frontier coding subscription under $20
- Claude Code pricing guide 2026
- Codex vs Claude Code comparison
- ChatGPT Work vs Codex — shared quota explained
- Token budget planning for production agents
Sources
- Tibo Sottiaux (@thsottiaux) — August 25, 2026 announcement
- Codex pricing — usage limits table
- Community discussion on X and Hacker News, August 25, 2026 (~115K views on primary thread)
Limit policy reflects Tibo Sottiaux's public post of August 25, 2026. Pro exemption duration is described as "upcoming months," not a permanent guarantee. Confirm your tier and meters in ChatGPT Work or Codex Settings before long agent runs. Follow @explainx_ai for the next policy move.