Day 22 of 2800:00:34left to ship today
What did Codex do to you today?
Tibo promised a real improvement or a full reset every day for 28 days. Say what Codex did to you, and see whether any day answers it.
One sentence. No account.
The 28-day promise
Tibo’s 28-day Codex promise, recorded day by day.
Tibo promised a meaningful improvement for most Codex/Work users, or a full reset, every day for 28 days.
Tibo · @thsottiaux · 4 Oct 2026 ↗Rantr is an independent, open record of what was promised, what shipped and what users say. Wins and problems are shown equally.
In their own words
What worked for others
Things Codex users on X say they did that helped. Each is one person’s report; Rantr has not tested it.
-
I’ve hit this exact same problem a few times. Pretty common right after a Codex upgrade. I fixed it by restarting my computer once or twice. Sometimes it works, sometimes it doesn’t. Just need a bit of luck. Pro tip: never update when you actually need the damn thing.
-
Hey The app is kinda bugging. Version 26.1021.31345 on Mac Silicon is not allowing me to do anything after reaching the usage refresh time. The only solution is to relaunch the app.
-
Hi Tibo, in ChatGPT's Windows Work mode, files attached in chat often go unread. Uploading to Library fixes it; web works fine. Astra seems worse than Sol. Anyone else? Is a fix coming? Thanks for the quota resets! The 5-hour limit feels smaller lately…😂😂
-
Changing reasoning effort still broke the cache for me (n=1). Codex CLI 0.164.0, GPT-6.1 Sol, exec resume: low to high, 7,168 of 57,950 input tokens cached (12%). High to medium 0%. Back to low 76%. Same effort after 1 min: 99.6%.
-
When codex stops to ask for input, the question is hidden and I have to press shift + left arrow to show it. Also, after a while, the question goes away and codex still continues working. Can you make it more prominent and user-friendly, please
-
比 grok bot 差太多了 1.github 无法拉仓库 授权了也没用，最后是通过建虚拟环境才解决 2.codex 的dots入口没了，需要浏览器输入 codex://dots才能出来 3.到底叫dot 还是dots，chatgpt页面叫： 你的dot 4.最好出个多dot 不然多个内容混一起有问题
-
The resume fix for long sessions is the one I'd feel most. I stopped trusting resume a while ago: when my context gets close to full I write a handoff note for the next window instead. Curious whether resume now replays the full transcript or a compacted state.
-
I had to spawn 6 conversation with Astra Ultra + fast speed + 4 sub agent/conversation, to be able to spend all my token on the same plan as you. How the hell are you spending this much token on normal usage?
Today · Day 22 · 26 Oct 2026
What Tibo shipped
22 of 22 days kept
22Tibo announces GPT-6.1 Terra, a model he says gives 2X more usage than 6.1 Sol across the API and subscriptionsImprovement
Tibo writes: "Day 22/ Terra is back! Our first model that is significantly better at writing, personality and nuanced understanding of requests. Packs a punch across the board, very good at computer use, and gives 2X more usage than 6.1 Sol across API and subscriptions." It was posted at 18:06 UTC on 26 October, within Day 22, quoting OpenAI's announcement: "GPT-6.1 Terra: delivering fast and capable computer use at a lower price."
Read the whole note
This is a new model release, not a usage reset and not a change to the stated plan limits. By Tibo's account, choosing Terra instead of 6.1 Sol makes the same allowance go about twice as far; the post does not say whether Terra is the default in Codex, which plans get it, or how it compares with Sol on coding. The 2X figure is his, not a benchmark.
To check it, select GPT-6.1 Terra in Codex or ChatGPT and compare how much of the allowance a similar task consumes against 6.1 Sol. Rantr has not independently verified this.
Replies to this announcement
Announcements are context. Your experience tells us what they actually changed.
02 → 04 · Who had it, the step, the status
What people describe, and where each one stands
Themes are what people describe, grouped by AI from Rantr experiences and collected X replies. Each one shows how many had it, what to do now, and whether an announcement answered it. AI groups; it does not verify claims.
Loading themes…
X thread: replies are found by search and each one is checked against its public post on X. X figures count only the replies read.
Read every collected reply to the promise
Sponsor
Sponsor the open record
One sponsor at a time, for seven days: a logo, one sentence and a link, shown directly below the experience composer. Sponsorship has no influence on votes or on what the record shows. No sponsor week is left to book in this campaign.
Community feedback
No experiences yet. Yours would be the first.
OpenAI release notes
What OpenAI published for this day. Answers about a note are counted apart from the vote on the day’s change.
Loading release notes…
How this campaign works
Writing is the contribution: you tell, in your own words, what the day’s changes did for you. AI groups related experiences into themes; a group shows experiences that resemble each other, not verified claims or representative polling. A theme’s status comes from AI linking a recorded announcement to it; the what-to-do line is written only from that announcement’s own words and checked against it before it is shown.
Rantr records source-linked updates; the community judges their usefulness. Our tracking window is 5 October–1 November 2026, using UTC days. The original post does not specify a timezone. Improvements and resets are counted separately, and a day can include both.
Historical votes (before input-only feedback)
One editable vote per update per browser, stored on Rantr. A 90-day essential cookie remembers this browser. Anonymous voting is not a verified-person poll. “Not tried” never affects the usefulness percentage. No fabricated votes or automatic failure verdicts.
Votes · Self-selected browser votes, not verified people or a representative survey. X replies do not enter this percentage. The crowd agrees with a day when its Useful votes outnumber its Not enough votes. Day records show a vote tally only once an update has 10 votes or more; smaller tallies stay in the open data.
What these numbers can tell you
Theme status · “Shipped” means an announcement was linked to the theme by AI, not that the problem is fixed for everyone. “Not addressed yet” means no recorded announcement was linked; it is not proof that nothing changed.
Recorded days · A day is kept when an improvement or a full reset is recorded.
Sources · Announcements, X replies and Rantr votes are separate sources. Votes are self-selected browser responses, not a representative survey. JEV identifies themes; it does not verify whether an experience happened. Read the original sources and compare both positive and negative experiences. X evidence covers only the replies Rantr has read: found by search, each one checked against its public post on X. It shows when they were read and how many. A reply not read is not a reply that does not exist. Imported replies never count as votes.
Release notes · Publication dates are not proof of availability for every account. Release notes do not automatically mark a campaign day as kept.
Community feedback · Related experiences are grouped automatically. Connections may be imperfect; they do not prove a fix. One experience per browser per day. “Same here” counts browsers, not verified people.
X replies · Replies to the original conversation. These do not judge a shipped feature or count as votes.
About Rantr. Rantr is where people say what a product did to them, anonymously, and see who feels it too. When the same thing keeps coming up, the pattern becomes visible to everyone, its makers included. Go to Rantr →
