Update — later September 29, 2026: Tibo quoted the usage thread and wrote: “That was yesterday, today is DevDay. And it's all good news.” The first post’s “tomorrow” was his timezone, not IST’s next calendar day. Reopen + unnamed extras are DevDay (September 29), not a separate September 30 SKU day. He still did not answer grandfathering, 30-day notice, or “is this $400 of API?” (it is not).
September 29, 2026 — Tibo Sottiaux posted the unglamorous note first: new $200 ChatGPT Pro sign-ups reopen, and how usage is calculated changes with them. His line: if you do the math, it nets out at half the dollar in API spend compared with the old Pro $200 plan. From the US evening of September 28 that was “tomorrow”; from India on the 29th it looked like the 30th until he corrected it.
That is not a gift of $400 of API. Several quote-tweets inverted the sentence. Half the printed API dollars is a tighter coupling to today’s (lower) list prices, plus whatever else OpenAI changed in the formula. He then argued you should still get more work done than on Pro a month ago, because models got cheaper and better, the 5-hour cap stays dead, and extra subscription features on DevDay will not eat the meter.
This is the reopen of the September 10 Pro sign-up freeze, after Astra-era 4x limit reports and the Sol/Luna 50% API cut. Read the formula before the keynote merch.
TL;DR
| Question | Direct answer |
|---|---|
| New $200 Pro? | Reopens DevDay, September 29, 2026 (US “tomorrow” on the first tweet; he later said today is DevDay). |
| Usage? | New calculation. Half the old plan’s API-dollar equivalent. |
| Twice the quota? | No. |
| 5-hour window? | Will not reintroduce, per Tibo. Weekly bucket when you want. |
| Why still “more work”? | Efficiency + passing through API price cuts (Sol/Luna 50% this month). |
| Existing 20x seats? | Not specified. Do not assume grandfathering. |
| $500 plan? | Not in this post. See the Pro Max $500 rumor. |
| Secret extras? | DevDay, features that do not draw usage. Still unnamed. |
What Tibo actually wrote (four claims, not one)
Paraphrase of the September 29 thread, not a screenshot rewrite:
- Reopen new Pro $200 subscribers the next US calendar day from that first tweet (DevDay).
- Change the usage calculation so the included allowance nets to half the API dollars of the old $200 Pro.
- Why you should not riot (his argument):
- (a) Will not bring back the 5-hour limit.
- (b) Over time the sub should deliver more work and higher quality as models get efficient and API prices fall.
- (c) They refuse to inflate list prices (or hide cuts in discounts) just so the sub looks like a huge API subsidy. GPT-6 Sol and Luna at 50% of previous price is the example. Long run: buy API as needed; the gap between sub and API per dollar should shrink.
- (d) Tomorrow: more things on the subscription that do not draw usage. Not named.
- He wanted this out before DevDay “big announcements” so the usage change is not buried in good news.
Signed “Codexingly” — same person who runs Codex and ChatGPT product comms for OpenAI.
The sentence everyone got backwards
Davey’s read: “$200 Pro will now get usage equivalent to $400 of API use?” That would be double the old API-dollar print. Tibo said half.
| Phrase | Meaning |
|---|---|
| Old Pro $200 | Whatever API-dollar equivalent OpenAI used internally / in marketing math before this change |
| New formula | Same $200 sticker, about 50% of that old API-dollar number |
| Not said | Token counts, Astra vs Sol mix, or a public calculator |
Independent testers have long argued maxed Pro 20x looked like thousands of dollars of list-price API (one widely cited SemiAnalysis-style ceiling was on the order of ten thousand-plus API dollars a month for a fully hammered seat). Those figures were never an OpenAI invoice. They are why a lab that cuts API prices 50% and still prints “$14k of value” looks like it is lying or subsidizing. Tibo’s (c) is the escape hatch: stop using inflated list prices as the scoreboard.
Mechanical identity (the charitable math):
If you mark usage at list price, and Sol/Luna list prices fell 50%, then the same tokens now print as half the API dollars. The headline can be a repricing of the meter, not a second 4x token cut.
Uncharitable math: they also reduced included tokens, and the 50% API-dollar line stacks on Astra weeks (Astra is still $10/$50 vs Sol $2/$10). Tibo did not publish a worked example with Astra vs Sol mix. If your Pro week is Astra agents, do not assume Sol’s price cut saved you.
Until OpenAI shows tokens, model mix, and the formula, treat “half API dollars” as confirmed as a claim and unconfirmed as a token table.
Worked example (illustrative, not an invoice): suppose an old Pro week was marked as $4,000 of list-price API. Half is $2,000 of today’s list. If that week was all GPT-6 Sol after the 50% cut, $2,000 at new Sol rates can be about the same tokens as $4,000 at old Sol rates. If that week was all Astra, $2,000 of Astra list is far fewer tokens than $4,000 of old mixed list. The formula change is neutral-to-good on Sol/Luna-heavy Codex, and painful on Astra-max ChatGPT. Tibo’s “more work than a month ago” only holds if you shifted mix or models got cheaper on the SKUs you actually hit.
Codex and ChatGPT share the same Pro seat in Tibo’s world. Burning Astra in the IDE and in the chat app double-counts against one weekly number. The July 5-hour removal already made it possible to dump the whole week on Monday. Keeping that property while reprinting half the API dollars is how OpenAI reopens the front door without restoring a rolling window that hid how fast a single agent session can empty the tank.
Follow-up replies still unanswered. Corey asked about current subscribers and claimed OpenAI T&Cs require 30 days notice. Moinul asked grandfathering. Paul still asked if this is $400 of API (no). Deepanshu asked for Astra 50% cheaper. Tibo’s second tweet only fixed timezone / DevDay. It did not confirm notice periods, legacy 20x math, or an Astra list-price cut. Screenshot your usage page; do not treat a quote-tweet as a contract.
Why reopen now (capacity timeline)
| Date | What happened |
|---|---|
| Jul 12, 2026 | 5-hour cap temporarily removed; weekly bucket only |
| Sep 5 | Astra full rollout + banked reset |
| Sep 7 | Reported 4x tighter Astra limits for heavy users |
| Sep 10 | New $200 Pro sign-ups paused — Pro was the highest per-user strain |
| Sep 22 | Sol/Luna 50% API cut |
| Sep 29 (US evening Sep 28 first tweet) | Tibo: reopen “tomorrow,” new usage math, no 5h return; later: today is DevDay |
| Sep 29 DevDay | Reopen + unnamed non-metered extras + keynote |
Pausing new Pro was the narrow capacity valve. Reopening it without bringing back 5-hour throttles means the weekly number has to do all the rationing. Halving the API-dollar print is how you reopen the door without restating “unlimited Pro.”
Current 20x subscribers and mid-cycle changes
Corey’s reply asked whether changing amounts mid billing cycle is even allowed, and whether legacy 20x should be kept.
What we know: Tibo did not say existing subscribers are grandfathered, pro-rated, or cut tonight.
What we will not do here: legal advice. Plan changes on SaaS are a contract question (OpenAI terms, your region, consumer law). Chinese-company comparisons are not a source. If you are mid-cycle and the usage page moves, screenshot before/after and ask support — that is ops, not a courtroom.
The Claude Max 20x lawsuit is the other lab’s multiplier vs weekly cap fight. Do not paste it onto OpenAI. Different vendor, different alleged facts. The lesson for buyers is the same: 5x/20x labels are not weekly token banks unless the help article says so.
“I’ll just go back to Anthropic”
That is a routing decision, not a morality play. If your work is scoped coding at volume, Sonnet 5.5 vs Opus 5.5 and what a Claude Code task costs are the actual bills. If you need Codex + ChatGPT in one login, the Pro reopen still matters even at half API dollars. Switching labs because a meter formula changed is rational; switching because of a quote-tweet is not.
The $500 rumor vs this post
EZRA: “Sounds like the $500 plan will end up being the old $200 plan.” That is a hypothesis, not Tibo’s text. The Pro Max $500 leak is still unconfirmed. A halved API-dollar $200 plus a future $500 would recreate a subsidized top tier above the reopened Pro. Wait for a price card, not a vibe.
(d) — extras that do not draw usage — is the only new SKU-shaped hint. Could be storage, connectors, Codex UI, or DevDay toys. Could be nothing you care about. Do not pre-spend.
Same day, the ChatGPT Models grid labeled GPT-5.6 Sol, Luna, and Terra as Unlimited* on Pro while GPT-6 stayed Expanded. Ian Nuttall mapped that star to (d); Tibo did not. Read the Unlimited* table separately — fair-use asterisk, likely chat-shaped, not confirmed for Codex hours.
What to do before you click subscribe
- If you were waiting on the pause: try DevDay / September 29. Expect worse API-dollar optics than the old 20x folklore.
- If you already pay $200: watch the usage page today. Compare weekly remaining and model mix (Astra vs Sol), not Twitter math.
- If you are Astra-maxing Codex: assume Sol’s 50% cut does not save you. Route cheap work to Sol/Luna; keep Astra for the jobs that need it (Astra vs Sol).
- If you live on API already: Tibo’s endgame is you. Sub vs API gap narrows. Price the API honestly with cache hits.
- Do not treat “half API dollars” as half of $200. The sticker is still $200. The included compute print is what moved.
- If you pool accounts or use unofficial “sub2api” bridges: that is still a ToS / fraud problem, not a usage-formula loophole. Official path remains Sign in with ChatGPT on your seat — see Codex limits vs sub2api. A halved API-dollar print makes pooling more tempting and more likely to get flagged.
DevDay coverage and Florida overlay: DevDay expectations. This usage post is the billing footnote Tibo did not want lost in the stream.
Related reading
- Pro Models grid: GPT-5.6 Sol/Luna/Terra Unlimited*
- ChatGPT Pro $200 sign-ups paused (Sep 10)
- Astra usage limits cut 4x (reported)
- GPT-6 Sol and Luna 50% API cut
- 5-hour limit removed (Jul 12)
- ChatGPT Pro Max $500 rumor
- OpenAI DevDay 2026
- GPT-6 Astra vs Sol
- Astra banked reset
- Codex limits vs sub2api
- Primary: Tibo on X (September 29, 2026)
Tibo’s posts are product comms, not a help-center rate card. Token tables, grandfathering, and unnamed extras land when OpenAI publishes them — tied to DevDay, September 29, 2026.