Oops! Something went wrong
[next-mdx-remote-client] error compiling MDX: Expected a closing tag for `<svg>` (6:56-6:61) before the end of `paragraph` 4 | - Large-scale human comparisons: 188,754 head-to-head pairs and 1,355,161 annotated votes across three independent questions (Preference, Coherence, Alignment). This emphasizes perceptual quality and prompt fidelity rather than proxy metrics. 5 | - Broad model coverage: 30 contemporary models (including multiple Claude, GPT, Gemini, Mistral, and other variants) evaluated on the same 500 prompts to enable direct cross-model ranking and ELO leaderboards. > 6 | - Reproducible pipeline and raw artifacts: models' raw <svg> markup is stored, repaired where possible, rasterized to 768×768 PNGs, and released with per-vote JSON so analysts can reaggregate or audit human judgments. | ^ 7 | - Prompt curation and licensing: prompt set mixes ~50 human-authored seeds with 450 samples drawn (and attributed) from a CC-BY-4.0 public dataset, selected for semantic diversity via embeddings and farthest-point sampling. 8 | More information: https://mdxjs.com/docs/troubleshooting-mdx
