Claude 3.7
A wheelie animation.
Source-labelled prompts, settings or separate runs are preserved without treating them as a ranking. Each real output counts once. No default/medium setting or source-default run is documented, so no timeline representative is guessed.
The author publishes a 12-second side-by-side recording of run 1 from three models. Video labels identify the model, high and s1. Only this model’s region is displayed; all 300 decoded frames match the original region pixel for pixel. The real one-second cover is not another work. The complete three-model recording remains folded in details. The two-sample design, timings and automated checks are author reports, not museum rankings or reproductions. The prompt forbids tools/iteration, but the author reports that grok-4.7 run 1 used a file-writing tool; this batch is not uniformly claimed to be strict tool-free one-shot generation. Model labels and attribution are source-reported, not authenticated. The date is UTC public posting, with generation time unknown. Both independent runs at the same setting count as works, not different reasoning levels. No default run or medium is documented, so no timeline representative is guessed.
Date precision / evidence: day / source-publication. Source publication: 2026-10-02. Generation date is not independently authenticated.
Original catalog label: high · run 1. A historical organizing label, not a score from a controlled benchmark.
A wheelie animation.
AnimationCommunity recordA conversation shared through Discourse AI.
An enhanced animation prompt with a surreal route, wings and shadows.