OpenAI spent DevDay selling an always-on agent. It spent the same week telling $200 Pro that Codex and ChatGPT Work would go from 20× Plus to 10× Plus on October 30. Then, on October 2, Tibo Sottiaux — Codex and ChatGPT — posted that GPT-6.1 Sol had a “slow start,” that it is “back to running at expected speeds,” and that a global reset lands tomorrow 10am PST for all paid ChatGPT accounts.
Read those as one product, not three announcements. Dots chats are free. The jobs a Dot starts are not. Those jobs hit the same Work / Codex meters OpenAI just halved. A weekend reset is the apology for the load spike. It is not a new SKU, and it is not a promise that October 30 will be generous.
TL;DR — three clocks, one bill
| Clock | What happens | Who it hits |
|---|---|---|
| Sept 29 (DevDay) | Dots ship; first Dot included on eligible Pro / Business Premium; extended deeper-work limits in month one | People who can actually log in |
| Sept 29 (same day) | Pro 500 at $500, Ultrafast, 25× Plus. Pro 200 reopens with a smaller formula | Anyone still on $200 |
| Oct 3, 10:00 AM Pacific | Tibo: global usage reset for all paid ChatGPT. Sol “expected speeds” after a two-day load spike | Plus / Pro / Business paid seats |
| Oct 30 | Pro 200 Work + Codex 20× → 10× Plus. GPT-6 Pro chat 200 → 100 / week. Price still $200 | $200 seats after the grace window |
| Dec 31 | One-time 62,500 credits (~$2,500) for grandfathered Pro 200 expire | Same seats |
The official subscriber email (quoted on Hacker News and in OpenAI’s own plan copy) is the source for the October 30 numbers. Tibo’s October 2 post is the source for the reset. We have not seen a model-card change attached to either.
What Tibo said on October 2 (and what replies got wrong)
Paraphrase of the October 2 post, not a screenshot dump:
- A global reset lands tomorrow, 10:00 AM PST, for all paid ChatGPT accounts.
- Apology for the slow start with GPT-6.1 Sol.
- Sol is back to expected speeds after a massive load spike in the first two days.
That matches his October 1 note that 6.1 Sol was “our most demanded model pretty much ever” on API and subscriptions, that ChatGPT and Codex were under heavy load, and that added capacity should approach ~2× yesterday’s speed. Demand is the compliment. The reset is the invoice.
It is not a new weekly cadence. Replies that asked “doesn’t everyone’s reset start tomorrow anyway?” are asking the right product question. OpenAI has used forced / early resets before (Tibo after the Sept 25–26 Codex / Work outage; the Astra quality postmortem reset). A global reset usually means: your remaining weekly bucket is replaced, not stacked. If you were hoarding 40% for Friday, you can lose that remainder when the clock hits 10:00 AM Pacific. Community posts after earlier forced resets described exactly that — a Sunday refresh with 80% left, then a new seven-day clock. Plan jobs after the reset if you were saving quota. Do not assume you keep both.
IST: October 3, 2026, 10:00 AM Pacific ≈ 10:30 PM IST (PDT). Tibo wrote PST. Check the Usage pane.
Dots: the launch looked messy because the launch was messy
The launch post still stands: GPT-6 Astra, own cloud PC and browser, 4,000+ plugins, first Dot included, chat does not burn ChatGPT, Codex / Work starts do. That accounting did not get kinder this week.
What did pile up:
- Live demos failed on the official stream (audio cut, a second demo also failed). OpenAIDevs posted a remake: “dots demo. Now... with better WiFi.” Same SKU. We already flagged that.
- The CFO said Muse on CNBC while talking about Dots — mix-up, not a rename.
- EEA / Switzerland / UK first-wave access has been reported as delayed. Official language remains “eligible markets.” If you are in those regions, do not budget a Dot for Monday.
- Free and Plus are still not the first wave. Always-on Astra is a Pro-and-up object.
- OpenAI promised extended limits in the first month for deeper Dot work. Month one ends in the same window as the Pro 200 cut. That is not a conspiracy. It is two calendars that rhyme.
Proactive research is still read-only. A Dot that only talks is cheap. A Dot that starts coding or Work is a Codex customer wearing a round face. If you wire Dots into an engineering org, you are buying another way to spend the meter OpenAI just cut.
Pro 200: the $200 sticker stayed. The 20× did not.
This is the part people mean by “halving usage on the $200 plan.” It is not a rumor. It is the DevDay plan change we first tracked as Tibo’s “half the API dollars” reopen before the email put integers on it.
From the subscriber note, as quoted in public threads:
Starting October 30, 2026, your included usage in ChatGPT Work and Codex will decrease from 20 times to 10 times the ChatGPT Plus allowance. Your GPT-6 Pro limit in Chat will decrease from 200 to 100 messages per week. Your price remains unchanged and you’ll keep your current limits through October 29, 2026.
Also in that note:
- They will not bring back the 5-hour limit.
- Dots are included; conversations and work the Dot does directly do not draw the ChatGPT allowance (Work / Codex starts still do).
- Pro 500 is the upsell: highest usage, 25× Plus, Ultrafast.
Existing eligible Pro 200 seats: old limits through October 29, plus a one-time 62,500 usage credits OpenAI values at $2,500, shown in Settings → Usage, expire December 31, 2026. New Pro 200 sign-ups after the reopen sit on the new (smaller) formula immediately — that is why “reopen” and “cut” landed on the same day.
The Hacker News thread title that stuck is “Pro 200 is fine, please don’t improve it.” That is the builder read. OpenAI’s read is: models got more capable and more expensive per useful hour, so the printed 20× was a subsidy they will not keep printing. Both can be true. Your job is to re-price the seat before October 30.
| Seat | List | Work / Codex vs Plus (stated) | Ultrafast | Dots (eligible markets) |
|---|---|---|---|---|
| Plus | $20 | 1× | No | No (first wave) |
| Pro 200 (after Oct 30) | $200 | 10× | No | First Dot included |
| Pro 200 (grandfathered through Oct 29) | $200 | 20× | No | First Dot included |
| Pro 500 | $500 | 25× | Yes | First Dot included |
Confirm the live grid on chatgpt.com/pricing. The Pro models grid still matters: last-gen 5.6 can look Unlimited* in chat while GPT-6 is Expanded. A Dot that starts Astra Work is not living on that star.
Why this week looks bad (without the doom caption)
Capacity. 6.1 Sol is the mid-ladder workhorse OpenAI wanted everyone on. “Most demanded ever” plus a two-day load spike plus a speed apology is what a successful price point looks like from the inside of a GPU queue.
Packaging. They shipped an always-on agent that is allowed to start the expensive surfaces, then halved those surfaces on the volume $200 seat, then invented a $500 seat for people who refuse the cut. That is coherent finance. It is a worse story than “Dots included.”
Trust. Forced resets, demo remakes, and a CFO name-swap are individually small. Together they read as a company that can ship faster than it can operate. Builders already learned this in July (banked resets, 5-hour cap removal) and in September (Astra quality + reset). The pattern is: break or overload → Tibo thread → reset. Useful. Not a strategy you can put in a quarterly plan.
Geography. An always-on agent that is late in the EEA/UK is not “global personal AGI.” It is a US-first beta with a DevDay keynote.
None of that means Dots is fake. It means you should not staff a workflow on “included deeper work” past the first invoice cycle.
What to do this weekend (and before October 30)
- After 10:00 AM Pacific on October 3, open Settings → Usage. Confirm the reset hit. If you had leftover quota you needed, it is probably gone. Start the jobs you postponed.
- Meter Dots as Codex. If the agent is starting Work / coding tasks, log those like any other Codex burn. Chat is the decoy meter.
- Re-run the $200 math. 10× Plus + 100 GPT-6 Pro messages / week + a $2,500 credit that dies on New Year’s Eve is a different product from 20×. If Codex is 90% of why you pay $200, compare API Sol (6.1 at $2/$10) and Pro 500 before you auto-renew on vibes.
- Do not buy Pro 500 for Ultrafast you will not use. Ultrafast is a speed SKU. If you needed 20× and not 300 tok/s, you are being upsold a different axis.
- Keep a second harness. When Sol is slow or a reset lands mid-sprint, Claude Code / API routing is the boring backup. Launch-week load is a vendor-risk item, not a personality flaw.
Related reading
- OpenAI Dots launch (access, meters, remake)
- Pro $200 reopen: “half the API dollars”
- Pro models grid: Unlimited* 5.6 vs Expanded GPT-6
- Ultrafast and Pro 500
- GPT-6.1 Sol pricing
- Astra quality postmortem + reset
- Sept 25–26 Codex / Work outage reset
- DevDay 2026 hub
Primary: Tibo Sottiaux on X (Oct 1 load / Oct 2 reset) · OpenAI Pro 200 subscriber email as quoted in public threads (HN 49901067) · Introducing Dots · Introducing GPT-6.1 Sol · chatgpt.com/pricing
Reset timing and Pro 200 integers are as of October 2, 2026 public posts and the circulated subscriber email. Confirm Usage and the pricing page before you change a seat. A global reset is not a price cut and not a new model.