← All benchmarksAI engineering
Claude Code vs Codex on the same ticket: both 14/14 hidden tests
How do Claude Code and Codex compare on the same small coding ticket?
On a CSV-export ticket with 9 acceptance criteria, 3 visible and 11 hidden tests, Claude Code finished in 36 s (about $0.20) and Codex in about 50 s; both passed all 14 tests. One run each on one small ticket.
Results
| Claude Code | 36 s, 14/14 |
|---|---|
| Codex | ~50 s, 14/14 |
How it was measured
Same prompt, one attempt each, fresh copy of the repo.
claude -p "Implement TICKET.md" ...
codex exec --sandbox workspace-write "Implement TICKET.md" ...Why
A well-specified ticket with tests makes the tool matter less.
When this does not apply
One run; not a benchmark of the models.
Environment
MacBook Air (Apple silicon, 8 cores), macOS. Measured 2026-10-12.
Source: shorts/demo-ticket/template in my content-engine repository. Treat any single run as a direction; re-run it on your hardware.
By Josué Barros, senior software engineer (full-stack, data and AI).