This Claude Code review is built on sixteen builds, not impressions. Anthropic's coding agent ran its two strongest models, Claude Opus 5.5 and Claude Fable 5.1, on the same eight briefs every other agent on this site received, one attempt each, nothing edited. Verdict first: Claude Code is the most thorough builder we have tested, and the most expensive in time and tokens.
| Brief | Finished | Wall time | Output tokens | Files | Source size | Runtime errors | API-equivalent cost | Attempts |
|---|---|---|---|---|---|---|---|---|
| SaaS landing page | Yes | 43m 20s | 144k | 1 | 132 KB | 0 | $7.15 | 1 |
| Analytics dashboard | Yes | 1h 21m | 285k | 1 | 115 KB | 0 | $13.63 | 2 |
| Balls in a spinning hexagon | Yes | 40m 11s | 211k | 1 | 31 KB | 0 | $8.73 | 1 |
| First-person shooter | Yes | 2h 30m | not recorded (stopped at the 150-minute limit) | 11 | 251 KB | 0 | — | 1 |
| Mobile finance app | Yes | 1h 8m | 295k | 1 | 115 KB | 0 | $21.06 | 4 |
| One-lane MOBA | Pending | — | — | — | — | — | — | 1 |
| Open brief | Yes | 1h 14m | 466k | 1 | 178 KB | 0 | $16.65 | 1 |
| Kanban board | Yes | 1h 40m | 446k | 1 | 113 KB | 0 | $19.98 | 2 |
What is good about Claude Code?
It checks its own work, and it shows. Claude Code runs the model in a loop: write, start a local server, open the page in a headless browser, look, fix. On the landing page, Opus 5.5 read files 59 times and ran 28 shell commands, about a dozen of them driving a browser; on the 3D shooter it ran 118. The result is builds that arrive polished and complete, with no runtime errors logged in our recordings of any finished build.
It also builds ambitiously. Claude Code's builds include working demos inside landing pages, annotated charts, haptic feedback in a phone app and an entire invented brand with sound for the open brief. Its two models diverge from each other more than Codex's do, so running both gives you two genuinely different options.
What is not so good about Claude Code?
Time, tokens and limits. Builds took 25 minutes to the full 150-minute limit; Opus 5.5 was still refining its shooter when the limit arrived. Every step re-reads the project, so a single landing page took about 10 million input tokens, mostly cheap cached reads but still billed. At list API prices, Claude Code builds cost roughly $7 to $25 each, against $0.54 to $7.27 for Codex.
On a subscription, that appetite collides with usage limits. Two maximum-effort sessions in parallel exhausted a five-hour window in about two hours, and several of our Claude runs needed a second attempt after a usage limit or an API outage. Every interruption is listed on its build; none was a model failure.
Which model should you run in Claude Code?
Claude Opus 5.5 by default. It built the richest results on most briefs, leads the public Terminal-Bench 4.0 leaderboard at 64.8%, and costs less per token than Fable 5.1, which Anthropic's model guide positions for long, demanding work. Fable 5.1 is faster and more restrained and, on Claude's pricing page, available on Pro only through extra usage credits.
How does Claude Code compare with Codex?
As the considered draft against the quick one. Codex finished every build in 7 to 25 minutes, in one pass, at a fraction of the cost; Claude Code took longer and built more. Our Codex vs Claude Code page shows all sixteen builds per agent, and the Codex review applies the same test to OpenAI's agent.
Is Claude Code worth it?
For client work, polished prototypes and anything you want right on the first try, yes, comfortably. For quick scaffolds and throwaway tools, a faster agent is the better use of your limits. Claude Code comes with Claude Pro, from $17 a month billed annually per Claude's pricing page; heavy users will want Max.
See if you can spot Claude Code's work: four invented brands, names hidden.
Take the blind test
Frequently asked questions
Is Claude Code good?
Yes. It is the most thorough coding agent we have tested: it built every brief we gave it from one prompt, tested its own work in a browser along the way, and produced the most elaborate results. It is also slow and token-hungry compared with OpenAI's Codex.
What are Claude Code's weaknesses?
Time, cost and usage limits. Builds took 25 minutes to the full 150-minute limit, cost roughly $7 to $25 each at list prices, and long sessions on a subscription ran into five-hour usage limits several times during our runs.
How much does Claude Code cost?
It comes with Claude Pro, from $17 a month billed annually or $20 monthly, and with the Max plans from $100 a month. It can also run on API billing, where Claude Opus 5.5 costs $4 and $20 per million input and output tokens.
Is Claude Code better than Codex?
For the first draft of an ambitious build, yes; for speed, cost and quick iterations, Codex wins. Claude Code with Opus 5.5 leads the public Terminal-Bench 4.0 leaderboard at 64.8% against 58.2% for Codex with GPT-6.