Evals

Each eval is its own chart with models on the x-axis; use the sidebar to pick a category, filter to specific models, or drop in a whole collection (Claude, popular open source, small models). Pin a model in the picker to keep it charted no matter the filter — everything else fades back so the comparison stands out. Switch to the compare card for a single shareable table: pinned model outlined, best score per row highlighted, missing figures dashed. Drag its column headers to put the models in the order you want — that order rides along in the URL. Either view copies to the clipboard as an image. Bars are colored by provider — a translucent bar marked with an asterisk (*) is vendor-reported, and a dashed-outline bar marked (△) is a community-reported median (submitted by contributors via the MCP server). Hover a bar for the variant, hardware, and source detail.