Recreating the Fault Once More, Free Reset with Caveats Pt. 5

I was already on the fence about keeping my ChatGPT Plus subscription after all that had happened. After the Opus 5.5 and Sonnet 5.5 announcements from Anthropic, I decided to let it lapse on November 16. I had cancelled earlier and found no reason to renew it. Claude’s lineup covers more ground, its subscription does more work for the same $20, and Codex has become slow.

Neither ChatGPT nor Codex gave a clear indication of when the subscription would die. My usage window, now thrown off again by a banked reset, said I had until sometime on November 20. When the subscription closed on the 16th, my access went with it. The window never showed when I would lose access either. It was a repeat of what happened to me before, except this time I knew it was coming. It is a fault in the Codex interface.

Versatility is the real deal-breaker here. Even on the $20 tier, Claude runs from Haiku to Opus. If I want the best model to solve a problem once and for all, Fable 5.2 is there despite its price, on usage or on Max. I have already shown that climbing that ladder on a Claude subscription is possible, and when the job is minimal, Haiku is ready. I can’t say the same with Codex. The offerings’ strong suit is affordability in API pricing, not performance.

Then there is the work per subscription. One tracker ran the same sessions on both $20 plans in October and roughly measured 71 million weekly tokens on Claude Pro with Opus 5.5 against 52 million on ChatGPT Plus with GPT-6.1 Sol, though it notes a token is not the same amount of work on every model. As of November 16, Wokey’s tracker puts the top tiers further apart: Claude Max 20x returns about 34 times its fee in API value, ChatGPT Pro $200 about 16. Third-party tools tell the same story. Anthropic prohibits using Claude subscriptions in other harnesses, while OpenHands’ own documentation says Plus and Pro subscribers can use Codex models on theirs. I take that as a sign Claude is far more subsidized. Neither company documents the subsidy, so take it with a grain of salt.

I also think OpenAI is shooting itself in the foot by giving out too many resets. Tibo, Codex’s lead, promised 28 days of either a daily improvement or a full reset. Claude hands out compute and calls it complimentary, which gives a better outlook on what the subscription is actually worth. Codex is so unpredictable by comparison that I cannot tell what a subscription is worth. The announcement came from Tibo’s own account on X, formerly Twitter, not from OpenAI’s official account or website.

Codex has also become slow. My own DIY projects took roughly 45 minutes to finish implementing a single step, and I do not recall a Claude run that long. That a tool this slow does not even offer auto-resume on a usage window makes it a damning experience, while Claude Code can continue on its own after a session limit resets.

This is not a question of brand loyalty, but of the reliability of a tool and the new billing model that comes with it. Codex was a great tool when it offered top-notch lineups, and had OpenAI kept that competitive edge, I would have kept it. If Anthropic pulls a similar stunt, I will stop my subscription immediately. As it stands, with only a monthly subscription I can start at any time, I have no problem pulling the plug on a GPT subscription.