Skip to content
GlanceHall
  • Builds
  • Models
  • Compare
  • How we test
Get the next round
  • Builds
  • Models
  • Compare
  • How we test
  • Journal
  • About
Get the next round

We don't rate. We let you see.

contact@glancehall.com

Watch

  • All builds
  • Models
  • Head to head
  • Journal

Trust

  • How we test
  • About
  • FAQ
  • Contact

Legal

  • Privacy
  • Terms
  • Cookies
© 2026 GlanceHallIndependent. Not affiliated with Anthropic, OpenAI or any model vendor.
Journal/Coding agents

What can Codex do? Watch it build 8 things

What can OpenAI Codex do? Watch it build eight different things from one prompt each: a website, a dashboard, a phone app, a kanban board, a physics toy, a 3D shooter, a MOBA and an invented brand.

By GlanceHall · Published October 22, 2026 · 2 min read

What can Codex do? The quickest honest answer is a demo, so here are eight. OpenAI's Codex agent, running GPT-6 Astra, built each of these from a single prompt with nobody touching the result. Watch them play, then open any one and use it.

GPT-6 Astra · extra high reasoning · Codex CLIAll of GPT-6 Astra’s builds
SaaS landing page
Open the saas landing page
Analytics dashboard
Open the analytics dashboard
Balls in a spinning hexagon
Open the balls in a spinning hexagon
First-person shooter
Open the first-person shooter
Mobile finance app
Open the mobile finance app
One-lane MOBA
Open the one-lane moba
Open brief
Open the open brief
Kanban board
Open the kanban board

Eight things Codex built with GPT-6 Astra, one prompt each. Click a title to open the live build.

What did Codex build in these demos?

One of each of the things developers most often ask an agent for.

  1. A marketing website with a working monthly-to-yearly pricing switch, an FAQ and a sign-up dialog. 10 minutes.
  2. An analytics dashboard that computes its KPIs from a fixed dataset, with sorting, filters, search, pagination and dark mode. 10 minutes.
  3. A physics simulation of twenty balls in a spinning hexagon, written from scratch. 8 minutes.
  4. A 3D shooter in three.js with waves, reloading and a game-over screen. 20 minutes.
  5. A phone finance app with a four-step send-money flow that updates the balance. 9 minutes.
  6. A one-lane MOBA with minions, towers, an AI enemy hero and four abilities on Q W E R. 25 minutes.
  7. A website for a product it invented: PEEP, a $28 ceramic bird whistle. 13 minutes.
  8. A kanban board with drag and drop between columns, undo, filters and a detail panel. 12 minutes.

How does Codex do it?

By reading, writing and running, in one long turn. Codex hands the model a folder and a shell. In our logs, each build was a single agent turn of 10 to 30 shell commands: writing the files, then running checks before finishing. The command we used, with every flag, is on the methodology page, and OpenAI documents Codex in its launch post.

Industrial machinery with robotic arm in a modern manufacturing facility.
Photo by Freek Wolsink on Pexels.

What can't Codex do, on this evidence?

Surprise you. Every build is complete and correct, and every build looks like it came from the same tidy studio: warm neutrals, serif headings, a gentle line of encouragement in the interface. When we ran the same briefs with GPT-6.1 Sol inside Codex, several builds came back nearly identical to Astra's. Codex also stops at the code: it does not buy a domain, host the site or connect a real payment backend unless you set that up.

How do you try Codex yourself?

Copy a prompt from any brief page, run it in Codex, and compare your result with these. OpenAI lists Codex among the features of its Pro and Business plans, and Plus subscribers can use their allowance in participating tools through Sign in with ChatGPT, per OpenAI's Help Center. To see what Codex built with the cheaper model, read the Codex projects gallery.

Four MOBAs, two built by Codex. Play detective before you reveal the names.

Take the blind test
Detailed view of machinery in an operational glass factory in Dar es Salaam.
Photo by Keegan Checks on Pexels.

Frequently asked questions

What can Codex do?

Codex is OpenAI's coding agent: it reads a project, writes files, runs commands and checks its work. From one prompt each it built eight working projects for us, from a marketing site to a 3D game, in 7 to 25 minutes per build.

What can't Codex do?

On our briefs it did not take creative risks: its designs follow one tidy house style and its two models often produced the same idea. It also does not set up hosting, domains or real backends unless you ask and give it access.

Is there a Codex demo I can try?

Every build on this page is live. Open any of them to use exactly what Codex produced, then copy the prompt from the brief page and run it yourself to see Codex work.

Which model does Codex use?

Codex runs OpenAI's models. We ran it with GPT-6 Astra and GPT-6.1 Sol at extra-high reasoning effort; this page shows Astra's builds, and Sol's are nearly identical.

Judge the builds yourself.

Eight briefs, four models, nothing edited. Watch them side by side or take the blind test.

Watch the buildsHow we test

Keep reading

Coding agentsCodex projects: 16 finished builds, from landing page to MOBASixteen projects built by OpenAI's Codex with GPT-6 Astra and GPT-6.1 Sol, one prompt each, no edits: websites, a dashboard, a phone app, a kanban board, a physics toy and two games.October 22, 2026Coding agentsOpenAI Codex review: judge it by 16 unedited buildsAn OpenAI Codex review based on evidence: sixteen builds from identical briefs with GPT-6 Astra and GPT-6.1 Sol, measured for time, cost and reliability, and compared with Claude Code.October 22, 2026Coding agentsIs Codex better than Claude Code? A per-brief verdict you can checkIs Codex better than Claude Code? Brief by brief, here is which agent did better on identical prompts, by what measure, and the builds behind every call so you can disagree.October 22, 2026