THE VERDICT
● Claude Pro for long unattended sessions. Claude Code ranks first among harnesses.
● ChatGPT Plus for breadth — Codex across web, CLI, IDE and iOS, plus GPT Image 2.
● Both changed this week. Neither change is a reason to switch on its own.
What changed on each
| Claude Pro | ChatGPT Plus |
|---|
| Price | $20 | $20 |
| Coding tool | Claude Code, ranked first | Codex, four surfaces |
| This week | Weekly limits drop 17% on 14 Sept | GPT-5.4 left Codex today |
| Image generation | Not bundled | GPT Image 2 included |
| Shared pool | Chat, Code and Cowork share one allowance | Metered per product |
THE SHARED POOL IS THE UNDERRATED DIFFERENCE
On Claude Pro, chat, Claude Code and Cowork all draw from one weekly allowance. A heavy Cowork week leaves less for coding.
ChatGPT meters separately, so image generation does not eat your Codex budget. That matters more after 14 September, when the Claude pool tightens.
Which one
| If you are... | Pick |
|---|
| Running long unattended refactors | Claude Pro, and watch the weekly bar |
| Doing varied work beyond code | ChatGPT Plus. Separate metering and bundled images |
| Pinned to GPT-5.4 | Switch Codex to API-key auth. It still has 5.4 |
| Able to run both | $40 removes two constraints and beats one $100 tier |
FAQ
Which is better for coding at $20?
Claude Code ranks first among agent harnesses for long autonomous sessions. Codex covers more surfaces and is better suited to pull-request-shaped work. Different jobs.
Are Claude Code limits dropping?
On 14 September weekly limits settle at 25 percent above the pre-May baseline, which is 17 percent below the current promotional level.
Can I still use GPT-5.4?
Through an API key, yes. Codex sessions authenticated with a ChatGPT account lost access today.