OpenAI doubles Pro plan usage limits, according to Theo's video

OpenAI doubled usage limits on all of its Pro plans effective October 29, according to Thibault (Tibo), who announced it on X, as Theo reported. The change covers the $100, $200, and $500 tiers and the business premium plan. Theo argues that the $200 plan is mostly being restored rather than expanded, and that the real gains are on the other tiers.

The restoration is central to the story. On September 28, OpenAI changed the $200 tier so that, by Theo's math, its API-equivalent value came to about half of the old plan's. Theo said the doubling brings the $200 plan back to roughly where it was a month earlier. Today's change restores the $200 plan but also doubles what the $100 and $500 plans give you, so those two tiers get a real increase. Theo said the $500 plan now offers about 50x the usage of the Pro tier, which he thinks you can probably only get through by using fast and ultra-fast modes heavily.

Theo's dollar estimates put the $500 plan at around 10.5 thousand dollars of usage and the $200 ChatGPT Pro plan at about 4.3k. He compared that with a $200 Claude plan, which he said gets 12 grand in usage, or six grand after a 50% cut on Fable limits. He noted that no Fable model was worth using over Opus at the time of recording, so the cut matters less in practice. He also said dollar value at API prices is a poor yardstick, because per-token pricing ignores how many tokens a model uses to finish a task and how caching changes the cost.

On per-task cost, Theo cited Artificial Analysis figures. The OpenAI model at its X high setting scored 52 on the Artificial Analysis Intelligence Index, against 48 for Opus 5.5 at medium. Opus 5.5 medium cost $2.45 a task, while the OpenAI model cost $0.76. He said the gap narrows once subscription pricing is factored in, and that at high settings the two come out about even, at roughly 5 cents and 4 cents a task.

Theo also pointed to blind reviews run through a skill that has agents from several model families review the same change. Opus preferred Opus reviews 66 times out of 83, and the OpenAI model preferred Opus reviews 54 times out of 83, which Theo put at 65%. The median cost was $1.65 for an Opus review and 36 cents for an OpenAI review. Theo estimated the expected API value of a $200 Claude plan at about $3,300 for Opus, and the $200 ChatGPT Pro plan at about $1,000 for the OpenAI model. That works out to about 205 reviews on Claude against about 290 on ChatGPT Pro. Theo said the gap is much smaller than the sticker price suggests, and that Opus 5.5's higher capability makes it the better value.

Theo also flagged resets as a factor that changes the comparison. He said OpenAI reset Codex limits 11 times in the last 30 days. From April 16 to October 31, he counted 52 resets in Codex and 14 in Claude, with none for Claude in the previous 30 days. He said that when resets are switched off in Adam's charts, the OpenAI limits look much worse and the two platforms look closer.

Chat and Work usage now also counts against Codex limits, which previously they did not. Theo said he thought this affected maybe 30 people, and that the doubling should more than cover it. He said the outrage over that change doesn't make much sense.

Theo's remaining complaint is the model itself. He said nothing OpenAI has released yet can go head-to-head with Opus 5.5, which he described as the best model available at the time of recording.