vLLM Semantic Router The Decision Box Gallery

Find one figure in a crowded masterpiece. Box questions on Decision 3 (d3-turbo and d3‑flash), live on the free Decision API.

How the search works

  1. d3-turbo: one request. One Box question on the whole work; d3-turbo answers it directly, with its own Box. No tiles, zoom or edge checks. Below 10% confidence means not found.
  2. d3-flash: four steps. I. Tiles: one yes / no question per tile, read at full resolution, and one for the whole work. II. Zoom: the likeliest tile, enlarged by a quarter. III. Box: one Box question on that tile, answered by 65 yes / no sub-questions on a zooming grid. IV. Edge checks: does an edge of the box cut the figure off? Cut edges move out, for up to six rounds.

The confidence is the model’s own estimate that its box overlaps the true one by at least half. Every request is live from your browser; nothing is precomputed. Compare sends both at the same moment and shows each model’s requests and answer time, measured in your browser.

Your own picture

Choose a picture and say what to find.

Run details

No search yet.

The works on view