Scope
BP Eval is an independent, read-only index of publicly released benchmark results. It does not run models or represent model providers, and it is not affiliated with LiveBench.
The release timeline indexes eleven frozen public snapshots. Model variants are collapsed to the strongest displayed result by default and can be expanded from the leaderboard controls.
Source
Raw task scores and category definitions come from the official LiveBench website repository. The latest source release contains 28 model snapshots across 23 tasks grouped into seven categories.
Source files are stored locally and frozen by date so readers can reproduce displayed values even after the primary leaderboard changes.
Aggregation
Each category score is the arithmetic mean of valid task values assigned to that category. Overall is the arithmetic mean of the displayed category averages; selecting a subset recalculates Overall across that subset.
Displayed values are rounded to one decimal place. Published cost views are shown only when a matching source cost table exists.
Limitations
Benchmarks are snapshots, not universal product rankings. Prompt settings, task selection, provider changes, and evaluation revisions can affect results. Consult the primary source before making consequential decisions.