THE VERDICT
● Claude Code for long unattended sessions. Deepest hooks, subagents and dynamic workflows, and it ranks first among harnesses.
● Codex for cloud, pull-request-shaped work across web, CLI, IDE and iOS.
● Tomorrow: GPT-5.4 and GPT-5.4 mini leave Codex for ChatGPT sign-in users. They stay available via API key.
Head to head
| Claude Code | Codex |
|---|
| Bundled with | Claude Pro $20 | ChatGPT Plus $20, also Go $8 |
| Harness ranking | First on hooks, subagents, workflows | Second, on cloud PR autonomy |
| Surfaces | CLI, desktop | Web, CLI, IDE, iOS |
| Buy more when capped | Max tiers at $100 and $200 | Yes on Plus. Free and Go cannot |
| Model change tomorrow | None. API pricing rises, consumer unaffected | GPT-5.4 leaves for sign-in users |
WHAT DECIDES WHETHER EITHER WORKS
Research on agent pull requests found reviewer engagement predicts merges better than model quality does. Large diffs merge less. Force pushes hurt.
Linear telemetry showed teams tripling weekly pull requests while total development time rose. Capping diff size in your instructions will do more than switching harness.
Which one
| If you are... | Pick |
|---|
| Running multi-hour refactors | Claude Code |
| Wanting work handed back as a PR | Codex |
| Pinned to GPT-5.4 in automations | Switch to API-key auth before tomorrow |
| Already paying for one | Use it. The gap is smaller than the switching cost |
FAQ
Which is better?
Claude Code ranks first among agent harnesses for long autonomous sessions. Codex leads on cloud pull-request-shaped work and ships across four surfaces. They target different jobs.
What changes on 31 August?
GPT-5.4 and GPT-5.4 mini leave Codex for anyone signed in with a ChatGPT account, replaced by GPT-5.6 Terra and Luna. Both remain available via API key.
Does Sonnet 5 repricing affect Claude Code?
No. That is API pricing. Claude Code stays bundled with Claude Pro at $20.