10 sources · 10 ingesting · 15,707 observations

Sources

Every published number traces to an immutable raw snapshot from one of these sources. Trust rank breaks dedup ties when two sources report the same result; each source’s default verification status is applied unless the data says otherwise.

Source	Adapter	Trust	Default verif.	Health	Fresh	Observations	Snapshots
Epoch AI Benchmarking Hub epoch.ai/benchmarks	epoch-ai	90	independent	healthy	0.1 h	1,159	2
Artificial Analysis artificialanalysis.ai	artificial-analysis	85	independent	healthy	0.1 h	2,651	1
LiveBench livebench.ai	livebench	85	independent	healthy	0.1 h	252	3
ARC Prize Leaderboard arcprize.org/leaderboard	arc-prize	80	independent	healthy	0.1 h	1,226	3
LMArena Leaderboard Dataset huggingface.co/datasets/lmarena-ai/leaderboard-dataset	lmarena	80	community	healthy	0.1 h	8,492	3
SWE-bench Leaderboard www.swebench.com	swebench	80	community	healthy	0.1 h	365	8
Terminal-Bench Leaderboard www.tbench.ai/leaderboard	terminal-bench	75	community	healthy	0.1 h	206	3
Scale Labs labs.scale.com/leaderboard	scale-labs	70	independent	healthy	0.1 h	257	13
Vals AI www.vals.ai	vals-ai	70	independent	healthy	0.1 h	1,036	11
LiveCodeBench Leaderboard livecodebench.github.io	livecodebench	65	independent	healthy	0.1 h	63	2

Health derives from recent ingestion runs (3+ consecutive failures = failing); Fresh = hours since the last successful run finished. Sources marked never-run are wave-2 adapters not yet ingesting. See methodology for how verification status qualifies every score.