This Codex review is built on evidence rather than impressions: sixteen builds made by OpenAI's Codex agent from the same eight briefs every other agent on this site received, with GPT-6 Astra and with GPT-6.1 Sol, one attempt each, nothing edited. Verdict first: Codex is the fastest, cheapest reliable builder we have tested, and the least adventurous one.
| Brief | Finished | Wall time | Output tokens | Files | Source size | Runtime errors | API-equivalent cost | Attempts |
|---|---|---|---|---|---|---|---|---|
| SaaS landing page | Yes | 9m 08s | 45k | 1 | 73 KB | 0 | $0.69 | 1 |
| Analytics dashboard | Yes | 7m 26s | 36k | 1 | 67 KB | 0 | $0.54 | 1 |
| Balls in a spinning hexagon | Yes | 7m 54s | 43k | 1 | 25 KB | 0 | $0.65 | 1 |
| First-person shooter | Yes | 13m 19s | 61k | 3 | 65 KB | 0 | $1.03 | 1 |
| Mobile finance app | Yes | 13m 46s | 48k | 1 | 57 KB | 0 | $0.83 | 1 |
| One-lane MOBA | Yes | 19m 41s | 67k | 3 | 85 KB | 0 | $1.12 | 1 |
| Open brief | Yes | 10m 51s | 40k | 1 | 65 KB | 0 | $0.71 | 1 |
| Kanban board | Yes | 14m 31s | 53k | 1 | 64 KB | 0 | $0.83 | 1 |
What is good about Codex?
It finishes, quickly, every time. All sixteen builds completed in one uninterrupted agent turn, in 7 to 25 minutes, with no runtime errors in our recordings. Every feature each brief listed is present, from a monthly-to-yearly price switch to drag-and-drop with undo on a kanban board. The finance app moves $42.50 and updates the balance correctly in both builds; the dashboards compute the right totals.
It is also economical. At list API prices, GPT-6.1 Sol's eight builds cost about $6.40 in total and GPT-6 Astra's about $31. Across all sixteen runs, 94 to 95% of the input tokens were served from the prompt cache, which is why the bills stay small.
What is not so good about Codex?
Variety, ambition and robustness under pressure. The two models it runs repeatedly produced the same idea twice: the same landing page typefaces, physics demos with the same name, shooters with the same door, invented products that both help you use your phone less. The designs are tidy and safe. In the games, the worlds are flatter than the Claude versions, and three of the four GPT heroes died at least once under our scripted input, as some Claude heroes did too.
How does Codex compare with Claude Code?
As the quick draft against the considered one. Claude Code took 25 minutes to two and a half hours per build, ran up to 118 shell commands checking its work, and cost about $7 to $25 per build at list prices, for richer and more distinctive results. On the public Terminal-Bench 4.0 leaderboard, Claude Code with Opus 5.5 scores 64.8% and Codex with either GPT-6 model 58.2%, at a far lower token cost. Our per-brief verdicts show where each wins.
Which model should you run in Codex?
GPT-6.1 Sol. It produced builds very close to GPT-6 Astra's at a fifth of the token price, and ties Astra on the public leaderboard. Astra was faster on two briefs and its shooter player survived where Sol's did not; neither is worth five times the cost for building.
Is Codex worth it?
For prototypes, internal tools and anything you will iterate on, yes, without much hesitation. For a polished build you want right the first time, pair it with Claude Code or plan to add the personality yourself. OpenAI lists Codex on its Pro plans, and you can see every build it made for us on the GPT-6.1 Sol and GPT-6 Astra pages.
Four finance apps, two built by Codex. Pick the one you would trust with your money.
Take the blind test
Frequently asked questions
Is Codex good?
Yes, especially for speed and cost. In our tests Codex finished all sixteen builds, two models across eight briefs, in 7 to 25 minutes each with no runtime errors in our recordings, at $0.54 to $7.27 per build at list prices.
What are Codex's weaknesses?
Ambition and variety. Its builds are complete but conservative, its two models often produce near-identical work, and its games are plainer than the Claude versions.
Is Codex included with ChatGPT?
OpenAI lists Codex among the features of its Pro plans, from $100 a month, and of ChatGPT Business seats. Plus subscribers can use their allowance in participating tools through Sign in with ChatGPT. Codex can also run on API billing.
Is Codex better than Claude Code?
Codex is faster and cheaper; Claude Code produces more elaborate builds and leads the public Terminal-Bench 4.0 leaderboard, 64.8% against 58.2%. Our per-brief verdicts are on a separate page.