An interactive chart · 48 models · 30 benchmarks · 467 sourced scores
The Race for the Frontier
Each vertical line is a benchmark, ordered easiest and most-saturated on the leftto hardest on the right. Each coloured line is a model, crossing every benchmark at its score. Because benchmarks use different scales — percentages, Elo, pass@k — every axis isnormalised to its own 0–100%; hover a point for the real number and the exact setting it was measured under. Drag the time slider to watch models arrive and thefrontier climb.
One cell per model and benchmark: the best publicly reported score under the model's strongest generally-available setting (pass@1, no external tools unless the benchmark is agentic). Hollow points are reported rather than independently confirmed. Dashed segments bridge benchmarks a model has no sourced score on — the line is not measured there. Every point links to its source.
Every score, in full
The complete matrix behind the chart — 467 sourced scores across30 benchmarks (rows, oldest and most-saturated first) and48 models (columns, by release date), so you can see exactly what we have and where the gaps are: 32% of possible cells are filled.Italic scores are company-reported rather than independently confirmed. Hover a score for the exact setting and provenance; click it to open the source. Scroll the table both ways.