What it tracks
Benchmark Radar connects new benchmark releases with their primary evidence: papers, code, datasets, public activity signals, and reported use by model developers.
Official project information
About Benchmark Radar
Benchmark Radar is a daily-updated tracker and searchable library for public AI benchmarks, evaluation datasets, and the evaluations used in model reports.
How to use it
Use the Radar for recent releases, the Library for established evaluations, and Trends to examine how benchmark activity changes across the same 20 research topics used in Library.
Evidence policy
Records link back to public sources. Missing signals remain unknown rather than being converted to zero, and inclusion is not a quality endorsement.
Official links
Benchmark Radar website
Official GitHub repository
Recent benchmark data
Benchmark library data
Benchmark discovery sources
The Library combines recent primary-source releases with benchmark directories from BenchLM, LLM Stats, VerdictPal, Model Beats, Snorkel, Terminal-Bench, Harbor Hub, and the Harbor publisher on Hugging Face.
Catalog inclusion is discovery evidence, not a quality endorsement. Records retain source attribution. Versions and evaluation subsets are kept distinct; training mixes, integration tests and non-benchmark metrics are excluded from the supplemental catalog audit. Unknown release dates are left unknown. BenchLM and LLM Stats refresh with the daily pipeline; the other sources currently use the audited September 15, 2026 snapshot.