Fifteen million. That is Tibo's number, not a chart screenshot.
On August 12, 2026, OpenAI Codex lead Tibo Sottiaux (@thsottiaux) reminded everyone he had pledged a reset for every 1 million additional Codex active users until 10 million, then admitted the team blew past that cap and went silent. Tease: "Little surprise for you tomorrow."
On August 13 he closed it: "Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast."
explainx.ai's last milestone post stopped at 8 million combined Codex + ChatGPT Work users on July 14. This is the next chapter: Tibo confirmed 15M and a reset is landing. Everything else in the replies — pay-to-reset, Astra, a new banked-reset SKU — is still a guess until he or OpenAI ships it.
Update — August 16, 2026: The pay-to-reset rumor flagged below stopped being a rumor. OpenAI is now testing paid resets priced $5-8 on Plus and up to $80 on the $200 Pro plan — replacing the free banked resets this post describes.
Update — August 21, 2026: Codex kept growing. Tibo confirmed 20 million active users and credited every Codex + ChatGPT Work user a free banked reset — usable on your own schedule, not a forced immediate refill.
Update — August 30, 2026: Tibo shipped another paid limit reset plus harness fixes (compaction, memory, goals, subagents, MCP) — OpenAI claims 10–50% efficiency gains on the same quota. Details: Codex capacity boost and fix batch.
Update — August 22, 2026: That 20M banked reset ran into the same kind of delay pattern this post's own reset once avoided. Tibo set an 8pm PST delivery deadline for it, the deadline passed with many users reporting no reset, and Ed Zitron called it "absolute marketing masterclass" — promising a reset, then not delivering one — while separately arguing OpenAI has cut rate limits and is "outright denying it." Full detail in the 20M post's update section.
Update — September 5, 2026: This tease's promise came full circle. After GPT-6 Astra's September 3 launch, Tibo announced the rollout finished ahead of schedule and credited every Plus, Pro, and Business user a full banked reset the same day — a bigger, broader version of the reset mechanic this post first covered. Full story: Astra rollout complete — full banked reset for everyone.
TL;DR — what people are asking
| Question | Answer (Aug 13, 2026) |
|---|---|
| Did Codex hit 15M? | Tibo yes — company-reported active users; not audited |
| Codex-only or Codex + Work? | Unspecified in these posts. July's 8M was explicitly combined |
| What is confirmed? | 15M crossing + a usage reset landing within ~1 hour |
| What is rumor? | Pay-to-reset, quota-formula tweaks, new banked resets, Astra |
| What was the 10M pledge? | Reset every +1M Codex actives until 10M — then silence |
| Is this a plan upgrade? | No — another manual Tibo top-up, like July and Aug 9 |
| 5-hour window gone again? | Not claimed. It came back July 30 |
What is /fast? | Codex CLI Fast service tier toggle — slash-command reference |
| Should I dump Ultra on toys? | No. Spend the refill on blocked real work |
What Tibo actually said
Two posts. Two facts. Do not flatten the thread into a product launch.
"I previously promised a reset for every 1M in additional active users for Codex, until 10M. We blew past that and have been silent since 10M. Little surprise for you tomorrow." — Tibo Sottiaux, August 12, 2026
"Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast." — Tibo Sottiaux, August 13, 2026
| Claim | Status |
|---|---|
| Codex crossed 15M active users | Confirmed (Tibo; "a bunch of days ago") |
| A usage reset lands in the next hour | Confirmed (Tibo) |
| Reset cadence was +1M until 10M, then silence | Confirmed (Tibo, describing his own pledge) |
| Pay-to-reset weekly quota | Not said |
| New banked reset inventory | Not said — July's banked-reset rollout is a different mechanism |
| Astra as the surprise | Not said |
| Permanent cap raise / 5-hour pause 2.0 | Not said |
explainx.ai read: The "little surprise" was the reset (and saying 15M out loud). If you held the tease for a model drop, you were trading on the gap Tibo left overnight — not on a second sentence he later posted.
Growth timeline — 1M to 15M
| When | Figure | Whose number | Notes |
|---|---|---|---|
| Feb 2026 | ~1M | Launch / community baseline | Desktop Codex era; 8M post used this as the start |
| Early Jun 2026 | ~5M weekly | Independent / community | Cited as ~6× since February — not a Tibo audit |
| Jul 12 | 6M active | Tibo | Same morning as the temporary 5-hour removal |
| Jul 14 | 8M active | Tibo | Codex + ChatGPT Work combined |
| Jul 21 | 10M | Community (Haider) | 1M → 10M in under six months; Tibo later said they blew past 10M and went quiet |
| Early Aug | 15M on charts | Community | Growth charts pointed at 15M before Tibo said it |
| Aug 13 | 15M | Tibo | "Old news" + reset landing |
Haider's public read: would not be surprised if Codex + ChatGPT Work had already crossed 15M. That is a combined-surface guess sitting next to Tibo's Codex-pledge language. Do not merge them into one MAU.
Metric hygiene: July 14's 8M was explicitly Codex + Work. The August posts say "for Codex" on the 10M pledge and "crossed that 15M" on the follow-up. OpenAI still has not published a help-center definition of active. Directional, not an S-1.
Same competitive pattern as July's Fable vs Sol limit week and July 16 three-vendor reset day: headline users, then a quota gift, then silence until the next spike.
What the community guessed (and what it isn't)
Overnight replies treated "little surprise" as a Rorschach test.
| Guess | Why it spread | Verdict Aug 13 |
|---|---|---|
| Free reset | Matches every prior Tibo gift | This is what landed |
| Quota tweaks | Sol still hungry after the ~18% efficiency pass | Unconfirmed |
| More banked resets | July shipped a manual button at 7M | Unconfirmed as today's gift |
| Pay-to-reset weekly quota | Power users want a price instead of a prayer | Unconfirmed — Tibo did not price a SKU |
| Astra | Separate OpenAI capability narrative; easy to bolt onto any tease | Unconfirmed — Astra debate is a different post |
Honest split: you can spend the reset today. You cannot budget a pay-to-reset line item or an Astra coding drop from this thread.
Practitioner guide — spend the refill, don't cosplay unlimited
What /fast actually does
Tibo's closer was not slang. In Codex it is a slash command.
| Command | Job |
|---|---|
/fast on | Turn on the Fast service tier for this model/thread |
/fast off | Turn it off |
/fast status | Print whether Fast is active |
Hidden /fast | Catalog has no Fast tier for the current model — pick another with /model |
Full map: Codex slash commands. Fast is catalog-driven. Persist it when Codex offers to save. Check /status before you assume the refill is burning Fast capacity.
Operator translation of "go /fast": use the gift on throughput, not on Ultra-for-fun. Fast is the lever that makes a one-hour window of extra quota feel like extra sessions.
Weekly cap vs 5-hour cap (do not mix July stories)
| Window | Mid-July | After Jul 30 | This reset |
|---|---|---|---|
| Weekly bucket | Repeated 100% Tibo top-ups | Still the plan governor | Refilled again (confirm in UI) |
| Rolling 5-hour | Temporarily removed Jul 12 | Restored Jul 30 | Not paused in today's posts |
| Banked reset button | Rolled out Jul 13–14 | Separate from auto weekly | Do not assume a new stack landed |
ChatGPT Work and Codex share one agentic pool on paid tiers — mode switcher does not clone quota. Details: Work vs Codex guide.
You already had a paid weekly reset on August 9 after the Tibo vs Boris harness-freedom thread. Today's gift stacks on that. It does not delete the 5-hour interrupt.
Don't burn the reset on fluff
| Do | Don't |
|---|---|
| Finish the migration / review / eval that hit the wall | Re-run last week's toy "write me a snake game" |
| Default Medium unless the task is novel | High / Ultra on boilerplate and reformatting |
| One Work deliverable or one Codex repo job that matters | Parallel fluff agents that cache-bloat the shared pool |
/fast on, then /status | Assume Fast is on because Tibo tweeted it |
Export artifacts locally if you remember Jul 16 $HOME wipes | Full-access unsandboxed marathons "because we have quota" |
Copy-paste check before you start:
1. Open Codex / ChatGPT Work usage UI — weekly bar and 5-hour bar
2. /status → confirm model, Fast tier, writable roots
3. /fast on (if the catalog exposes it) → /fast status
4. Pick ONE blocked job. Not three nice-to-haves.
5. Keep High/Ultra for architecture and security diffs only
What is still unknown
- Active-user definition. Session in 30 days? Weekly? Codex CLI vs desktop vs Work? Tibo did not attach a footnote.
- Who got the reset. July free-tier accounts sometimes lagged paid. August 9 was paid Work + Codex. Today's wording was "everyone" — verify your meter.
- How long until the next silence. The +1M-until-10M machine is retired. There is no published cadence past 15M.
- Whether 15M is Codex-only. Pledge language says Codex; July's 8M said both surfaces. Ask OpenAI, not a screenshot.
Tibo remains the public reset button. That is effective marketing and a bad SLA. Plan sprints on plan quotas, treat gifts as bonus compute.
Summary
Tibo Sottiaux confirmed Codex crossed 15 million active users and told subscribers a usage reset would land within the hour — after going quiet past his 10 million "reset every 1M" pledge. That is the news.
The rest of the thread is fan fiction until it ships: pay-to-reset, extra banked buttons, quota-formula tweaks, Astra. Use /fast if the catalog has a Fast tier, remember the 5-hour window is back, and spend the refill on work that was actually queued — not on proving you can empty a bar. Follow @explainx_ai when the next Tibo meter move lands.
Related on explainx.ai
- Astra rollout complete ahead of schedule — full banked reset (Sept 5)
- Codex hits 20M users — banked reset for everyone (Aug 21)
- OpenAI starts selling usage resets — up to $80 on the $200 Pro plan (Aug 16)
- Codex Multi-Agent V2 — delegation, GPT-5.5, and the Luna removal (Aug 16)
- Codex + ChatGPT Work hit 8M users (Jul 14)
- ChatGPT drops 5-hour limit — Tibo weekly reset (Jul 12)
- GPT-5.6 Sol ~18% longer + 5-hour returns (Jul 29)
- Sol thinking budget + banked resets (Jul 13)
- Fable 5 vs GPT-5.6 limit resets
- Claude, Codex, Cursor — limit reset day
- ChatGPT Work vs Codex — complete guide
- Tibo vs Boris — Aug 9 paid Work/Codex reset
- Codex slash commands — fast, model, status
Sources: Tibo Sottiaux on X (August 12–13, 2026 posts). Community growth charts and Haider's 10M / combined-15M comments are unconfirmed except where they match Tibo's own words.
User counts and reset timing reflect Tibo's August 12–13, 2026 public posts. Active-user definitions are not independently audited. Limit policies are temporary gifts unless OpenAI publishes a plan change — verify meters in-product before you schedule a production sprint.