Changelog

What changed: the tools for comparing designs, the models, and the skills. Versions use release dates. Unreleased changes are still being tested.

Unreleased

In progress · not in the public collection yet

Models. GPT-6.1 Terra added to the experiment setup through Codex, across all four briefs and samples. Public results are pending.
Skills. Hallmark v2 added to the experiment setup with its full reference package. Public results are pending.

October 27, 2026

Loading. Smaller JavaScript bundles and one combined stylesheet per page. Updated files get new URLs so browsers can cache them without missing changes.
Site. Renamed the homepage to Collection. Added this changelog, linked from the footer.
Fixed. The matrix map stays visible when you drag to the bottom. Dragging stops at the last row without cancelling navigation.
Loading. The matrix opens with a smaller dataset. Full attempt history loads when you open History; Stats still uses the full measurements.

October 23, 2026

Collection. 1,784 published designs across four briefs and 18 model rows. Coverage varies by model, skill, and sample.
Models. DeepSeek 4.1 Flash Base and 1017 have separate rows, preserving results from the earlier checkpoint. Reflection’s Beam is also included.
Skills in this snapshot. Frontend Design, Web Guidelines, UI UX Pro Max, Impeccable, Taste, Distinctive Websites, and Best Web Design, alongside Vanilla.

October 16, 2026

October 14, 2026

Models. Expanded Claude coverage. The public preview reached 15 model rows and 1,461 published designs.
Filters. Added the updated provider filters to the public preview, along with repaired design previews.

October 11, 2026

Collection. Published the first preview: 990 interactive designs across four shared frontend briefs.
Privacy. Shared design previews and experiment measurements. Custom briefs and execution logs stayed private.