Three ways to compare price

  1. Token prices depend on input, output, and caching. A lower list price per token does not establish a cheaper completed task. 6:05
  2. Task cost also depends on how many tokens the model needs to solve the problem. Different reasoning settings can change that consumption. 6:05 6:28
  3. Subscription cost per task additionally depends on the effective allowance or discount supplied by the plan. This can change the API-price ranking. 7:34

The cited benchmark comparison

  • In the chart Theo describes, 6.1 Soul on max uses about 68,000 tokens, roughly comparable to Fable on medium/high. Soul on x-high uses fewer tokens than the recent Anthropic configurations shown except Opus 5.5 low. These are chart-specific observations. 6:28
  • Soul x-high scores 52 on the Artificial Analysis Intelligence Index versus 48 for Opus 5.5 medium. Theo says Opus medium costs more than twice as much per task while scoring lower in this comparison. 6:59
  • The transcript renders the Opus price as “$245” and the Soul price as “76 cents.” The precise Opus amount is ambiguous in the supplied text; these notes retain the stated relative comparison rather than silently inserting a decimal. 6:59

Model names such as “6.1 Soul” and “Fable” follow the supplied transcript. The transcript is retained verbatim for checking ambiguous wording.