Home / Reviews / ChatGPT Codex

Baseline run · Benchmark 1: SlotKeeper

ChatGPT's Codex built a working booking app, but needed help getting it online

Oct 25, 2026 · ChatGPT Plus features (a Pro account limited to Plus) plan · 10/10 checks · ~20 min

slotkeeper-maple-yoga.netlify.app

Open the live app (ChatGPT Codex's app, opens in a new tab)

Live app at slotkeeper-maple-yoga.netlify.app, captured Oct 25, 2026, opening with “Find your balance. Make time for you.” Slot dates and spots left are today's, not the test day's.

From the test day

Select any screenshot to open it full size.

§01 Verdict

On a ChatGPT Plus-level plan, Codex built SlotKeeper in under 8 minutes with its own tests passing, then stopped: it couldn't deploy on its own. After 2 fix prompts it was live on a temporary link about 20 minutes after the prompt; once the tester made a free Netlify account, it deployed permanently. It passed all 10 checks with 3 minor bugs. Score: 82.5/100, as the general-AI baseline, not ranked with the specialists.

§02 How it went

1. We submitted the same prompt at 07:20:40 EDT. Codex finished building at 07:28 with its own tests passing, but didn't deploy: it said it needed hosting access.
2. Fix prompt 1 asked it to put the app online with free hosting. Netlify rejected the test email address and Cloudflare stopped at a CAPTCHA.
3. Fix prompt 2 asked for a way online without an account. At 07:40 the app was live on a 60-minute temporary link with a Neon Postgres database. That link has since expired.
4. Later that morning the tester created a free Netlify account and token, and one deploy-setup prompt asked Codex to deploy permanently. It did, in about 6 minutes, and moved the data across. All 10 checks passed again on the permanent URL.
5. The run used a clean setup with no connected apps, plugins or personal accounts.

Deploy

Codex deployed to Netlify's free plan with a Netlify-managed Postgres database and moved the data across itself. The tester only created the free Netlify account and an access token. Its first public URL was a 60-minute temporary link that needed no account.

Cost

$0 usage. It runs inside a ChatGPT subscription (Plus-level features); hosting is on Netlify's free plan.

Stack

React and Vite front end, an Express API, Postgres (Netlify-managed).

Model

GPT-6 Astra, default reasoning

App URL

slotkeeper-maple-yoga.netlify.app (ChatGPT Codex's app, opens in a new tab) (recorded on Oct 25, 2026; test data only. Netlify free plan; free usage limits apply.)

§03 Checklist

1. Landing page lists the 3 seeded slots with spots left.PASS
2. Booking with a valid name and email succeeds, a confirmation shows, and spots go down by 1.PASS
3. A missing name or invalid email is rejected with a clear message.Clear message from the server; an empty form gets only the browser pop-up.PASS
4. Data persists after a hard refresh (redeploy not tested).After a hard refresh. Redeploy claimed by Codex, not re-tested by us.PASS (refresh)
5. A full slot (capacity 1, after one booking) shows "Join waitlist" and assigns a waitlist position.PASS
6. The same email can't double-book the same slot.PASS
7. The owner can sign up and log in, and logged-out /admin redirects to login.A second owner sign-up is blocked; Codex created the first owner itself.PASS
8. The owner can create, edit and delete a slot.PASS
9. Cancelling a booking promotes the first waitlisted person automatically.PASS
10. Stats line counts are correct, and it all works on the public deployed URL.Stats tiles matched the computed counts.PASS

§04 What broke

• MINOR

  An empty booking form shows only the browser's own validation pop-up.

  To reproduce: Submit the booking form with every field empty.

• MINOR

  Owner sign-up closes after the first owner, and Codex created that first owner account itself.

  To reproduce: Open the owner sign-up page once an owner exists.

• MINOR

  Codex left deployment credentials, including the database connection string, in a temporary folder on the test machine.

  Repro steps not recorded.

§05 Caveats

NoteBaseline run, not ranked with the specialists: it ran on Oct 25, a day after the last specialist, on a ChatGPT Plus-level paid plan rather than a free plan.

NoteBug score under review: our rubric gives 12 of 15 for 3 minor bugs; the scorecard recorded 11. Under our own rule the total would be 83.5. We show the recorded 82.5 until it's checked; it doesn't change anything above it on the board.

NoteIt needed 2 fix prompts and a human-made Netlify account to get online. The five specialists needed no fix prompts and published from their own Publish button.

NoteHuman cleanup minutes weren't timed. The human steps that were recorded are on the receipt.

NoteDesign is a single tester's estimate.

NotePersistence across a redeploy is Codex's claim; we didn't redeploy it ourselves.

NoteIts first public link (a 60-minute temporary deploy) has expired. The score uses the permanent Netlify URL, re-checked 10/10.

All open questions: corrections log →

§06 Compare

Quick compare:vs Bolt.newvs Base44vs v0vs Lovablevs Replit Agentvs Cursor

Visit ChatGPT Codex (plain link, no affiliate code)