User measures what $20 buys on Claude, Codex and Grok
A test posted on Thursday puts Claude Pro at 115 million tokens a week against 29 million on Codex Plus, and rates the cheaper plan higher.
An account posting as shownotover published a comparison on Thursday of what a $20 subscription buys on Claude Pro, Codex Plus and SuperGrok, run with each model at its highest setting. It reported 115 million tokens a week on Claude Pro with Opus 5.5, against 29 million on Codex Plus with GPT-6 Sol.
The account rated the results in the other direction. It scored Codex Plus 7 out of 10 and Claude Pro 5, with SuperGrok on Grok 4.7 at 6. Those scores are its own judgement of the output, not a benchmark, and the post gives no task list.
A second test points the same way on price. An account posting as TheHunterBohm said it gave Opus 5.5 and GPT-6 Sol the same prompt, and logged 12.7 minutes and $2.86 for Opus against 11.4 minutes and $1.09 for Sol. That post had 219,383 views.
David Heinemeier Hansson said the latest Rails agent evals still show OpenAI ahead, despite what he called a good jump from Opus 5.5. He singled out Luna Max as the surprise, at 18 per cent completion for $11, which he said was not far off GPT-6 Sol.
None of these is a controlled test. A single prompt, a weekly limit read off a dashboard and a score out of ten measure different things, and two of the three accounts are judging their own output. What they agree on is that the cheaper option is closing the gap.