10 reference pages

Benchmark library

Task-and-grader profiles for the evaluations carrying the main capability claims in this research.

Generation and reconstruction · 2024

A benchmark for reconstructing a scientific plot as executable plotting code from its image.

Generation and reconstruction · 2025

An end-to-end benchmark combining data, a question, an answer, chart code, and annotated visual evidence.

Generation and reconstruction · 2026

A chart-to-code benchmark built from real source data, multi-panel tasks, and iterative refinement.

Chart reading and critique · 2025

A realistic chart question-answering benchmark with conversational, hypothetical, fact-checking, and unanswerable cases.

Professional chart reading · 2026

A deliberately difficult set of practitioner-authored professional chart-reading tasks.

Multiple charts and documents · 2026

A multi-chart scientific-figure benchmark whose name collides with a separate multilingual chart benchmark.

Languages and chart reading · 2026

A multilingual chart-question-answering benchmark distinct from the similarly named multi-chart scientific-figure dataset.

Multimodal evaluation · 2026

A multilingual pairwise-preference benchmark for multimodal judges that includes a chart-centric subset.

Interactive dashboards · 2026

A benchmark for reconstructing interactive Plotly Dash applications from screenshots, optional DOM, and interaction.

Interactive dashboard generation · 2026

A benchmark, browser executor, and human-calibrated judge for open-ended interactive dashboard generation.