My real-world usage test
Claude Code vs. ChatGPT Codex
Same prompt. Same video project. Two $20 plans. One used a quarter of its five-hour allowance. The other burned through all of it in under ten minutes.
One project, not a universal benchmark. Your usage may look different.
Watch the test
25% vs. 100%. That got my attention.
I dug into three possible explanations: model costs, growing conversations, and the limits themselves. The result changed which tool I reach for first.
What the comparison tells us
Same price does not mean the same amount of work.
Same monthly spend
I paid $20 for ChatGPT Plus and $20 for Claude Pro. That matched the spend, not the size of the two usage allowances.
Same job
Both tools got the same prompt and video project. I wanted to see how much of each allowance that work consumed.
Very different usage
Claude Code used 25%; Codex used 100% in under ten minutes. This is what happened in my session—not a promise for yours.
Model cost is only one clue
The video discusses token prices, but API pricing does not tell us how subscription limits are calculated.
Context adds up
Files, tool output, follow-ups, and previous work can make a tiny request much bigger than the message you typed.
A daily driver and a closer
My takeaway is not to pick a universal winner. I start with Claude Code and save Codex for the difficult jobs.
Claude Code
25%
of my five-hour allowance used
ChatGPT Codex
100%
of my five-hour allowance used—in under 10 minutes
These are percentages of two different allowances. This test does not measure identical token budgets, completion quality, or every kind of coding task.
The token-price caveat
The script names GPT-6 Astra at $10 input / $50 output per million tokens and Claude Opus 5.5 at $4 input / $20 output. Those are recorded claims from the video, not independently verified current rates. Even if API rates differ by 2.5×, that does not prove how either $20 subscription calculates its allowance.
Confirm current details with OpenAI pricing and Anthropic pricing.
My workflow now
Claude Code is my daily driver. Codex is my closer.
The better question is not “which one is better?” It is “which one should I use for this job?”
Step 01
Start everyday work in Claude Code
For routine building, I start with Claude Code. I do not turn up the thinking level unless the task actually needs it.
Step 02
Keep context useful, not endless
When I start unrelated work, I clear the conversation. If the current work still needs its context, I compact it instead. Save the important decisions and unfinished work first.
Step 03
Bring in Codex when the job calls for it
If the code still will not work, another model gets stuck, or the problem genuinely needs deeper reasoning, I bring in Codex. That is when I am willing to spend more of its allowance.
Getting started? Read my Claude Code guide or see my Premiere Pro workflow.
Compare your own workflow
Track the work, not just the bill.
A companion checklist for recording your prompt, model, context, allowance before and after, and whether the result actually worked.
Download checklist (.md)No email required. Companion checklist based on the video; not the original test prompt.
Which one burns through your limits faster?
Tell me in the video comments. I am curious whether your experience matches mine—or looks completely different.
More from Chad
Keep exploring
- Visit page
Claude Code Guide
Install Claude Code and build your first real tool in 10 minutes.
- Visit page
Connect Premiere Pro to Claude Code
Cut retakes, add zooms and graphics, mix audio, and inspect exports with plain-English prompts.
- Visit page
Stop Burning Lovable Credits
Edit your Lovable project for free with GitHub and Claude Code.