Doing the math on OpenAI halving Pro Codex allowance: it's not that bad
The author argues Pro's API-equivalent usage drops from about $14k to $7k a month, still close to Claude Max, and the real issue is that Astra 6 is not a strong enough model.
People are freaking out about OpenAI cutting the $200 Pro Codex allowance in half, but the math really isn't as bad as people make it sound. ChatGPT Pro was roughly ~$14k/month in API-equivalent usage. After the cut, that's around ~$7k. Claude Max is around ~$8k. That's barely a meaningful difference for the same $200. The actual problem is that Astra 6 is just bad model. Claude is smarter and more efficient, so you can simply get more out of the same amount of compute. If OpenAI announces a new model at Dev Day that fixes that, this comparison changes immediately. And people are comparing these subscriptions like ChatGPT Pro is just Codex. It isn't. You also get unlimited image generation, unlimited regular ChatGPT on a separate pool, Pro models that are much smarter than any Fable model for actually hard problems and can even be used with Codex, plus Finance, Health and a bunch of other services Claude doesn't have. They announced 20+ releases today, including a new bots mode that will probably have its own separate quota too. Claude's $200 plan is basically a huge bucket of model usage. OpenAI's $200 plan is an entire platform. And that's probably the whole point of the $500 plan. If all you care about is hammering Codex all day, that's the plan meant to compete for that audience. The $200 Pro plan is increasingly selling the whole ecosystem.
This post quotes another post: “OpenAI reopens $200 Pro subscription with a new usage calculation”