What now shares the allowance

  • ChatGPT and Work now count against the same limits as Codex; previously their usage was separate, according to Theo. Heavy users of those products therefore need to consider their combined consumption. 13:49
  • Theo expects ordinary chat and Work usage to consume much less than daily coding-agent use, so he thinks the doubling should cover typical combined usage. This is his expectation, not a measured result presented in the video. 13:49 14:16
  • He initially says Dots remains outside the shared limits, then immediately admits uncertainty. The video does not establish Dots' actual billing treatment. 14:16 14:40

How Theo chooses

  • For maximum review throughput, he presents Soul as an option for people willing to accept lower preferred-review quality for roughly 30–50% more reviews under his estimates. For the best model, he favors Opus and argues Claude's subscription value is closer to Codex than the API prices suggest. 12:58 13:25
  • His own division of work is OpenAI for computer use and verification, Claude for daily coding, and Grockbot through a separate subscription. 14:40
  • He welcomes the limit increase, while stressing that the $200 change restores the previous allowance. His remaining objection is model quality: he wants an OpenAI release that competes directly with Opus 5.5. 15:01 15:28