Loading blueprint versions...
Please wait while we gather all the unique runs for this blueprint.
Please wait while we gather all the unique runs for this blueprint.
Please wait while we prepare the detailed comparison.
This blueprint evaluates a model's ability to generate increasingly complex SVG illustrations. The challenges are organized into explicit difficulty tiers (easy, basic, intermediate, advanced, expert) and range from basic geometric shapes to detailed scenes, testing various SVG features like gradients, patterns, animations, and complex paths. This blueprint is designed to be a capability benchmark, not a bias test, so it does not contain any assertions.
--eval-method embedding for similarity analysis or --eval-method llm-coverage for rubric-based evaluation.