Update — July 29, 2026: Tibo reset Work/Codex limits again, ships Sol efficiency expected to last ~18% longer, and will restore the five-hour limit on July 30 — walking back this temporary pause. Full breakdown: Sol usage ~18% + 5h returns.
Update — July 24, 2026: Desktop ChatGPT Voice burns standard Work/Codex quotas for voice-triggered agents — same buckets as this limits guide.
Update — July 19, 2026: @ClaudeDevs extended Claude Code weekly limits 50% higher through August 19 — Anthropic's counter-move in the same retention week as Tibo's 5h cap removal. explainx.ai founder @goyashy publicly noted Codex signup in reply. Combined July 18 guide
Update — July 16, 2026: Tibo (@thsottiaux) investigated GPT-5.6 Codex file deletions — including
$HOMEwipes when full access bypassed sandboxing and auto review. Same day as limit reset day; do not spend fresh quota on unsandboxed runs: Codex $HOME deletion.
Update — July 16, 2026: Limit reset day — Anthropic refilled Claude 5h + weekly buckets; Codex users reported another same-day refill (Theo). Combined coverage: Claude, Codex, Cursor limits reset.
Update — July 15, 2026: Tibo announced 8M active users across Codex + ChatGPT Work — up from 6M on July 12 — with another usage reset and 5h cap still off. See 8M growth breakdown.
Update — July 13, 2026 (afternoon): Tibo denied GPT-5.6 Sol thinking-budget nerfs, promised ~10% more usage from inference savings, and rolled banked resets to 500k users — all Work/Codex users get one July 14 at the 7M milestone.
Update — July 13, 2026: New guide — ChatGPT Work vs Codex maps the three-mode split (Chat / Work / Codex), shared agentic quota, and cloud vs desktop sync gaps from OpenAI Help Center + r/codex.
The rolling 5-hour window is gone — for now.
On July 12, 2026, Tibo Sottiaux (@thsottiaux, OpenAI Codex) posted three launch-week fixes after GPT-5.6 Sol demand doubled traffic in 48 hours:
- Temporarily remove the 5-hour usage limit on Plus, Pro, and Business
- Make GPT-5.6 Sol more efficient — less quota burned per task
- Reset weekly usage to 100% within the next hour
Codex crossed 6M active users the same morning. Developers reporting in — including @andr3barroso — confirmed 5-hour meter disappeared while weekly balance refilled.
Same weekend Anthropic extended Fable through July 19. Two labs, same playbook: remove friction before subscribers churn.
TL;DR — Tibo July 12 updates
| Change | Detail |
|---|---|
| 5-hour limit | Temporarily removed · Plus, Pro, Business |
| Governance | Weekly cap only (during promo) |
| Weekly reset | 100% refill · landing within ~1 hour of post |
| Sol efficiency | Lower quota consumption · impact TBD |
| Codex MAU | 6M active users |
| Permanent? | No — Tibo said temporarily |
| Prior resets | Jul 10 ×2 · Jul 11 ×1 · Jul 12 ×1 |
Tibo's thread (July 12) — verbatim structure
"Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates:"
| # | Update |
|---|---|
| 1 | Temporarily removing the 5 hour usage limit for Plus, Business, Pro |
| 2 | GPT 5.6 Sol efficiency rollout — less usage per task, further impact quantified later |
| 3 | 6M active users · usage reset in the next hour — "Go do things" |
BleepingComputer (Jul 12, 8:44 PM): Codex and ChatGPT share a single usage pool — local messages + cloud tasks count together.
What changed vs before
Old dual-layer system
ChatGPT / Codex paid tier
├── Rolling 5-hour window ← hit this → hard stop
└── Weekly cap ← separate ceiling
Long Sol Ultra or multi-agent sessions could exhaust the 5h bucket while weekly quota remained — the top complaint in Tibo's replies.
July 12 system (temporary)
ChatGPT / Codex Plus / Pro / Business
└── Weekly cap only ← 5h gate removed
@andr3barroso tested live: "5-hour limit is gone, weekly balance reset to 100%."
Not unlimited — weekly still caps total frontier usage. OpenAI also promises Sol burns less of that weekly pool after efficiency patches.
Launch-week reset timeline — four interventions
| Date | Tibo action | Context |
|---|---|---|
| Jul 9 | GPT-5.6 Sol/Terra/Luna GA | OpenAI thread |
| Jul 10 | Reset Codex + ChatGPT Work ×2 in 24h | "demand surge unmatched" |
| Jul 11 | "Introducing... another usage limit reset" | ~30 min rollout · "never seen traffic increase so quickly" |
| Jul 12 | Remove 5h cap + weekly 100% reset | "didn't get everything quite right" apology thread |
Manual resets are OpenAI tradition — Tibo shipped similar May 15 and June 4 top-ups after outages. Launch week added structural change (5h removal) on top of quota gifts.
Why OpenAI scrambled — Tibo's admission
From the July 10–11 apology thread (summarized by The Decoder):
| Problem | Fix in flight |
|---|---|
| High-compute mode burned quota invisibly | Default settings + model picker adjusted |
| ChatGPT Work desktop overhaul hid chats/projects | Sidebar return next week |
| Codex deprecation fear | Tibo: "we love Codex and it is here to stay" |
| Multi-agent / plugin breakage | Patches deployed |
| Usage visibility | Clearer metrics + reset times promised |
5-hour removal is the fastest lever — no UI sprint required.
X reaction — surprise and skepticism
| Reply | Take |
|---|---|
| @andr3barroso | Verified 5h gone · weekly 100% · prefers weekly-only for long dev sessions |
| Anonymous | "Please permanently remove the 5-hour limit" — Tibo didn't promise |
| Anonymous | "5 hour limits made no sense when there's already a weekly quota" |
| Anonymous | "Wow, you remove the 5 hour limit, not just increased it" — resubscribe bait |
| Anonymous | "When will the 5 hour limit get implemented again?" |
Theo (@theo) earlier same week: burned $200K+ tokens on Sol — "a bit too easy to hit your limits on the $200 Codex Pro sub" — then praised OpenAI's generous resets. July 12 changes directly address that pain.
Competitive frame — same weekend as Anthropic
| Lab | July 12 move | Mechanism |
|---|---|---|
| OpenAI | Remove 5h · weekly reset · Sol efficiency | Tibo thread · 6M Codex users |
| Anthropic | Fable promo → July 19 · 50% higher Code weekly limits | @claudeai · 10.7M views |
@ClaudeDevs still runs 5-hour + weekly on Claude Code. OpenAI testing weekly-only on ChatGPT tiers is a product experiment — if retention holds, 5h may not return.
@theo on claudex: Sol in Claude Code harness — builders can arbitrage OpenAI weights + Anthropic scaffolding while both labs fight over limits.
What developers should do now
- Check /usage or Codex status — confirm 5h meter gone · weekly at 100%
- Burn Sol while reset fresh — Terra/Luna for cheap probes · Sol for hard tasks
- Avoid accidental Ultra — Tibo fixed defaults but verify model picker
- Plan around weekly only — longer sessions OK; still finite
- Watch Tibo on X — resets are manual, not calendar-predictable
- Compare $/task vs Fable — included Fable through July 19 vs Sol weekly pool
What to watch next
| Signal | Why |
|---|---|
| 5h cap returns? | Tibo said temporary — no end date published |
| Sol efficiency metrics | Tibo promised quantified savings |
| ChatGPT Work sidebar update | "Next week" per apology thread |
| Extension #4 on Fable | July 19 cliff vs OpenAI weekly-only experiment |
| Codex #31814 | Ultra subagent burn in native Codex — claudex workaround |
Related on explainx.ai
- GPT-5.6 Sol ~18% longer + 5-hour returns (Jul 29)
- ChatGPT Voice on desktop — GPT-Live quotas for Work/Codex agents
- Pliny 42K-word GPT-5.6 Sol Codex prompt leak — Tibo points to openai/codex (Jul 15)
- ChatGPT Work vs Codex — what actually changes
- GPT-5.6 Sol thinking budget — Tibo denies nerf, 10% usage bump
- Fable & GPT-5.6 limit resets — July 9–12 overview
- Claude Fable 5 + Code limits through August 19 — Anthropic counter
- Fable 5 extended to July 19 — prior promo terms
- GPT-5.6 vs Fable 5 comparison
- GPT-5.6 Sol in Claude Code — claudex
- Codex vs Claude Code verdict
- macOS blocks Codex as malware — XProtect false positive (Jul 2026)
- Claude Code vs Codex rate limit history
- Why AI companies want agent usage — token economics
Sources: @thsottiaux July 12, 2026 · BleepingComputer — GPT-5.6 limits relaxed · The Decoder — ChatGPT Work fixes · @thsottiaux July 11 reset
Limit policy reflects Tibo's July 12 posts and reporting as of July 13, 2026. Temporary removals can revert without notice — verify in your account usage panel before planning production sprints.