The earlier cut

  • Theo dates the earlier announcement to September 28. He says its changes to the $200 plan halved the allowance measured in API-equivalent dollars. 2:47
  • The announcement promised to remove five-hour limits and increase the amount and quality of work a subscriber could complete as models improved. Theo supports judging plans by completed work. 2:47 3:24
  • OpenAI's stated rationale was to avoid incentives to inflate API list prices to make subscription allowances look larger. Theo separately speculates that cheaper 6.1 Soul API pricing motivated the cut by increasing the compute delivered for a fixed dollar allowance. 3:24 3:46

The October 29 increase

  • The new announcement doubles limits for the $100, $200, and $500 Pro tiers and Business Premium, according to Theo's account. 0:24
  • Only the $200 plan had suffered the earlier cut. Its doubling restores the previous level; the $100 and $500 plans receive a net increase over their earlier allowances. 4:11
  • Theo describes the $500 plan as offering “50x usage compared to the pro tier”; the transcript does not clearly identify that baseline. He expects heavy fast/ultrafast use to be the likeliest way to exhaust it. 4:34

Reported API-equivalent allowances

  • OpenAI's $500 plan offers about $10,500 of API-equivalent usage, and its $200 plan about $4,300. These are comparison estimates, not cash credits or task counts. 5:03
  • Claude's $200 plan offers about $12,000 by the same measure. Theo mentions a model-specific 50% restriction for the model transcribed as “Fable”; even applying that reduction gives $6,000, above OpenAI's $4,300. 5:03 5:33