• A chart that counts extra usage resets makes OpenAI look substantially more competitive than one using the base allowances. Theo initially suspects the discrepancy comes from the new doubling, then identifies resets as the cause. 10:21
  • He reports 11 OpenAI resets in the preceding 30 days. Turning off resets in the comparison reduces OpenAI's apparent value. 10:21
  • Between April 16 and October 31, the cited data records 52 Codex resets versus 14 Claude resets. Theo says Claude had none in the preceding 30 days. These are historical counts in the video, not promised future allowances. 10:45 11:08
  • Even with reset accounting, the charts still describe Artificial Analysis benchmark work. Theo turns to his own code reviews to compare a task he actually performs. 11:08