September 29, 2026 — While Tibo was explaining that reopened $200 Pro usage nets to half the old API dollars, Ian Nuttall posted the plan Models grid: GPT-5.6 Sol, Luna, and Terra are Unlimited* on Pro. GPT-6 Astra / Sol / Luna are Expanded.
That is the missing give-back in the Pro reopen thread — if the star covers the surfaces you actually use. Replies already split: chat buffet vs Codex still metered. Tibo has not said this table is item (d) (“more things that won’t draw on usage”).
TL;DR
| Question | Answer |
|---|---|
| What’s new on the grid? | Pro: Unlimited* on GPT-5.6 Sol, Terra, Luna and GPT-5 Thinking Mini. |
| GPT-6? | Expanded on Pro (Astra, Sol, Luna). Plus: checkmarks, not Unlimited. |
| GPT-5.6 Sol Pro? | Pro-only check; not labeled Unlimited*. |
| Free/Go 5.6 Sol? | Dashes — not unlimited there. 5.6 Luna still checked on Free/Go. |
| Balances halved API dollars? | Plausible. Not Tibo-quoted. |
| Codex / GH-in-chat for hours? | Assume no until the footnote is public. |
| Same as August Free Luna unlimited? | No. That was Free/Go text chat. This is the Pro column. |
The table (what the screenshot actually shows)
Source: OpenAI ChatGPT plans Models UI, captured September 29, 2026 (user screenshot). Columns: Free, Go, Plus, Pro.
| Model | Free | Go | Plus | Pro |
|---|---|---|---|---|
| GPT-6 Astra | — | — | ✓ | Expanded |
| GPT-6 Sol | — | — | ✓ | Expanded |
| GPT-6 Luna | — | — | ✓ | Expanded |
| GPT-5.6 Sol | — | — | ✓ | Unlimited* |
| GPT-5.6 Sol Pro | — | — | — | ✓ |
| GPT-5.6 Terra | Limited Work/Codex desktop | same | ✓ | Unlimited* |
| GPT-5.6 Luna | ✓ | ✓ | ✓ | Unlimited* |
| GPT-5 Thinking Mini | ✓ | ✓ | Expanded | Unlimited* |
Expanded ≠ Unlimited*. GPT-6 stays on the metered / expanded allowance language OpenAI already used for paid tiers. Last-gen 5.6 mid stack is what they stamped Unlimited*.
GPT-5.6 Sol Pro is a separate row: Pro gets a check, without the Unlimited* label. Do not collapse “Sol” and “Sol Pro.”
Ian’s read vs Tibo’s (d)
Ian: OpenAI is giving *unlimited (fair usage) of those models to balance the usage-formula change. He later pointed at Tibo’s (d) — extras that won’t draw on usage — as why he thinks it is not “just unlimited in chat.”
What is actually in evidence:
- A UI table. Screenshot is real enough to transcribe; footnote copy is not in the image.
- Tibo (d) is still unnamed. Mapping it to this grid is inference.
- Rohan: “That’s just unlimited usage in Chat.” Ian disagreed, citing (d).
- newmike: connect a GitHub repo in chat and run hours — you learn unlimited is limited.
- Yash (@goyashy): the * is the worry.
- Talha: fair usage is the buffet star. Still a real upgrade if Sol and Luna xhigh stay open.
explainx.ai’s rule: UI label < help-center footnote < Codex rate card. We have the first. We do not have the second or third from Tibo.
Why unlimiting 5.6 (not GPT-6) is the cheap lever
GPT-6 Sol/Luna just took a 50% API cut. Astra is still $10/$50. Serving Unlimited* Astra on a $200 seat would recreate the SemiAnalysis-style subsidy Tibo said they will not print as fake list-price value.
Unlimited last-gen Sol/Luna/Terra* is how you:
- Reopen new Pro sign-ups without Astra-all-you-can-eat.
- Give power users a pressure valve that is not the flagship.
- Keep GPT-6 Expanded as the paid-shaped SKU.
Howaboua asked why they’d unlimit compute-heavy models. Relative to Astra, 5.6 Sol is the cheaper generation. Relative to GPT-6 Sol, 5.6 is the one they can afford to star. If your favorite is 5.6 Sol xhigh, you win on this grid. If your week is GPT-6 Astra Codex, you are still on Expanded plus the new API-dollar formula.
Jonas’s “5.6 Sol is kind of better than 6.0 Sol” is a preference, not a benchmark table. Run your coding eval. OpenAI shipped GPT-6 Sol as the successor at half the old Sol list. Both can be true: you like 5.6’s personality and 6.x is the SKU they’re metering.
Fair usage, Work/Codex desktop, and the August Luna story
Asterisk in consumer SaaS almost always means: no reseller farms, no 24/7 unattended loops, we can slow you down. Until the footnote is public, plan as if multi-hour agent + repo is not the unlimited SKU.
GPT-5.6 Terra on Free/Go is already limited in Work and Codex on desktop in this same table. That is a surface split inside one model family. Pro Unlimited* does not automatically delete that split. Chat vs Work vs Codex vs API are four bills.
August 7: Free/Go unlimited text chats on GPT-5.6 Luna, not Work, not API, not Codex. Today’s Pro grid is the paid analog for more of the 5.6 stack — still likely chat-shaped until proven otherwise.
If you were about to unsub Pro over half API dollars, try routing grunt work to 5.6 Unlimited* and keep GPT-6 Expanded for the jobs that need it. If Codex is 90% of your $200, wait for the star’s legal line or stay on API Sol/Luna pricing.
What to do this afternoon (DevDay)
- Open ChatGPT → Plans → Models. Confirm the Pro column still matches this screenshot after the keynote. UI ships faster than tweets.
- Read the * tooltip / footer. If it only says fair use, screenshot it.
- Run a timed Codex session on GPT-5.6 Sol vs GPT-6 Sol. If both decrement the same weekly bar, Unlimited* is chat theater.
- Do not assume grandfathering of old 20x math. Tibo still has not answered current subscribers (same post).
- Do not buy Pro because someone said 5.6 > 6. Buy it if Unlimited 5.6 chat* plus Expanded GPT-6 matches how you work.
How to route work if the star holds
Treat the grid as three buckets, not one “Pro is unlimited” slogan.
Bucket A — last-gen 5.6 with the star: draft, rewrite, long chat, Luna xhigh if that is still your taste, Terra if the table still lists it Unlimited* on Pro. This is the pressure valve after halved API-dollar math.
Bucket B — GPT-6 Expanded: Astra, GPT-6 Sol, GPT-6 Luna. Use when you need the current stack. Expect a meter. The 50% API cut does not make ChatGPT Pro Astra free.
Bucket C — Codex / Work / desktop / API: still separate. The same screenshot already shows Terra limited on Free/Go in Work and Codex on desktop. Pro Unlimited* does not rewrite that split by magic. Work vs Codex still applies.
If Ian is right and (d) is bigger than chat, you will see Codex remaining stop decrementing on 5.6 Sol. If Rohan and newmike are right, chat stays open and repo-connected hours still die. Run that experiment once and keep the screenshot. That is the whole post.
Plus vs Pro on the same screenshot
Plus already has checkmarks on GPT-6 Astra/Sol/Luna and on GPT-5.6 Sol. That is access, not Unlimited*. Plus GPT-5 Thinking Mini is Expanded. Pro is the only column with Unlimited* on 5.6 Sol/Terra/Luna and Thinking Mini.
If you pay Plus hoping this screenshot means 5.6 Sol buffet, it does not. The star is a Pro column label. Plus users still sit on the same weekly rationing OpenAI has been tightening since Astra 4x reports and the 5-hour window removal.
GPT-5.6 Sol Pro remaining Pro-only without Unlimited* is the other tell. OpenAI still wants a named Pro variant of last-gen Sol that is not the buffet SKU. Treat it as optional / high-effort, not as “the unlimited row.” If your picker still shows Sol Pro, assume it draws usage until the tooltip says otherwise.
GPT-5 Thinking Mini Unlimited* on Pro is the cheap overflow model in the same table. Useful if you want volume after 5.6 Sol is throttled under fair use. Useless if Mini cannot hold your coding context. Do not plan a week of Codex on Mini because the cell says Unlimited*.
What “Expanded” has meant in this product
OpenAI’s paid ChatGPT copy has used Expanded for higher caps, not no caps. After they killed the 5-hour Codex pulse, weekly remaining became the only consumer-facing bar most people watch. Expanded on GPT-6 Pro is consistent with that: more than Plus, less than a star.
That is why this screenshot can be good news and still match Tibo. He is reopening $200 while halving the API-dollar print. The grid then stars last-gen and expands current-gen. The $200 sticker does not become $400 of Astra. It becomes $200 of mixed SKUs, some of which look unlimited in chat.
If you are comparing labs, Claude Max multiplier fights are a different contract. Do not paste Anthropic lawsuit facts onto this table. Do paste the same buyer habit: screenshot before/after, record which model the usage page decrements, and do not pool seats (sub2api vs Sign in with ChatGPT).
What would falsify this post
This write-up is wrong if, after DevDay:
- The Pro column drops Unlimited* on 5.6 Sol/Luna/Terra.
- The * tooltip lists chat only and excludes Codex, Work, and connectors.
- A 5.6 Sol Codex session decrements the same weekly bar as GPT-6 Sol.
- Tibo names (d) as something else (storage, a new app, voice hours) and not this grid.
Until one of those lands, the honest headline is: Pro starred last-gen 5.6 (fair use unknown); GPT-6 stays Expanded; Codex is unproven. That is enough to route chat today. It is not enough to cancel API or buy Pro solely for agent hours.
How to read the X thread without getting played
Ian’s first post is the table. That is the load-bearing artifact. His later posts are interpretation: he wants (d) to mean more than chat, because Tibo said extras will not draw usage and refused to name them. That is a reasonable bet. It is still a bet.
Rohan’s “just Chat” is the conservative bet and matches August: Free/Go Luna unlimited was text chat, not Work, not Codex, not API. OpenAI has a habit of starring the cheap surface.
newmike’s GitHub-in-chat hours test is the operational test. If you only read marketing, you will discover the star after a long agent loop. If you run the test before you re-up $200, you will know whether Unlimited* is a chat cell or a compute cell.
Talha’s buffet line is the legal test. Every all-you-can-eat product keeps a fair use kill switch. The interesting question is where they set it: shared accounts and scrapers, or a dedicated Pro user grinding 5.6 Sol xhigh for a workday. We do not have that line in Tibo’s thread.
Howaboua’s “these models are expensive” is true vs Mini, false vs Astra. GPT-6 Astra Expanded on the same page is OpenAI saying the expensive one is still metered. 5.6 is last-gen. Thinking Mini Unlimited* is even cheaper overflow. The grid is internally consistent even if it feels weird if you still think 5.6 Sol is the frontier.
Jonas preferring 5.6 Sol to GPT-6 Sol is taste. GPT-6 Sol’s launch card is the SKU OpenAI is selling to API buyers. Preference and list price can disagree. Your coding eval decides which one you pin in Codex. The plan grid only decides which one looks unmetered in ChatGPT.
Subscribe or wait?
Subscribe today if you already wanted Pro for ChatGPT + Codex in one login, you can live on 5.6 chat for volume, and you accept GPT-6 Expanded plus half API-dollar optics.
Wait 24 hours if Codex is the product you buy. The star is not a Codex rate card. DevDay can still rewrite the tooltip.
Stay on API if you already price Sol/Luna with cache and you do not want a fair-use surprise. Tibo’s endgame for heavy users is still that you pay list.
Do not treat Unlimited* as permission to resell a Pro seat. That was already a ToS problem when the print was 20x folklore. A starred last-gen SKU makes pooling more tempting and easier to detect.
Related reading
- Pro $200 reopen: half API dollars, timezone, DevDay
- Pro sign-ups paused (Sep 10)
- GPT-6 Sol and Luna 50% API cut
- August Free/Go unlimited Luna text
- GPT-6 Astra vs Sol
- Astra usage 4x cut
- 5-hour Codex limit removed
- DevDay 2026
- Codex vs sub2api
- ChatGPT Work vs Codex
- Primary: Ian Nuttall on X · Tibo on X (September 29, 2026)
Models grid transcribed from a ChatGPT Pro plans screenshot on September 29, 2026. Unlimited-star is a UI label until OpenAI publishes the fair-use footnote and whether Codex shares it.