Claude Opus 5.5: Low vs Max on the Same Tesla Ad

Can more AI thinking turn the same Tesla photographs into a better commercial? I gave Claude Opus 5.5 the same Roadster brief in five fresh sessions, changing only the effort setting: low, medium, high, extra high and max. Watch all five complete 30-second ads, then compare the launches, wheel animation, editing, completion times and reported usage. Pick your favorite before the verdict. 00:00 One prompt. Five Tesla ads. 00:26 The shared brief and rules 01:00 Low effort 01:51 Medium effort 02:45 High effort 03:39 Extra-high effort 04:35 Max effort 05:31 Launch moments compared 05:51 The limits of the source photos 06:09 How long each run took 06:29 API-equivalent usage 06:51 My verdict 07:12 Which ad would you choose? HOW THE EXPERIMENT WORKED Each run used the same supplied Tesla Roadster photographs and reference frames, a fresh Claude Opus 5.5 session, and a 60-minute limit including rendering and verification. The ads were built with code, with their own sound, without a video-generation service. This is one attempt per effort setting. All five runs shared one computer, so completion times are observations, not a controlled speed benchmark. Dollar figures are API-equivalent usage reported for runs on an existing subscription, not five separate cash charges. The comparison edit, captions and narration were added afterward. Narration uses OpenAI Cedar. The car photography was supplied; animation, editing and sound were the model challenge. Independent experiment and unofficial Tesla concept ads. No affiliation with or endorsement by Tesla or Anthropic. Which effort level would you choose—and which shot convinced you? #Claude #Tesla #AI