Sources / livebench

LiveBench

healthydefault: independent

livebench.ai · trust rank 85 · adapter livebench

Observations

252

append-only score facts

Raw snapshots

immutable fetched payloads

Last success

2026-07-14 00:15Z

end of last succeeded/partial run

Freshness

1.4 h

hours since last success

License & attribution

The LiveBench repository is Apache-2.0; the site CSVs themselves are unlabeled and are surfaced as facts with attribution and a link. Newest questions are deliberately withheld under their contamination policy.

What we ingest

Per-release table CSV paired with its same-version categories JSON; task scores average to category means (their method). Each dated release is a separate benchmark version, and inference-config slugs stay separate evaluated systems.

Ingestion runs

last 3 of the run history

Started	Duration	Status	Key stats
2026-07-14 00:15Z	0.1 s	succeeded	docs 3 · promoted 0 · blocked 0 · scoreAdded 0 · unchanged 3
2026-07-14 00:08Z	0.4 s	succeeded	docs 3 · promoted 0 · blocked 0 · scoreAdded 0 · unchanged 3
2026-07-13 20:03Z	0.3 s	partial	docs 3 · staged 252 · promoted 0 · blocked 252 · scoreAdded 0 · unchanged 0

partial = run finished but some rows are blocked on identity review — nothing is published until every referenced model/agent/benchmark is resolved.

Raw snapshots

latest 3

Fetched	URL	SHA-256	Last re-fetch
2026-07-13 20:03Z	https://livebench.ai/cost_2026_06_25.csv	b4743ce3760a…	2026-07-14 00:15Z
2026-07-13 20:03Z	https://livebench.ai/categories_2026_06_25.json	dad300ad1865…	2026-07-14 00:15Z
2026-07-13 20:03Z	https://livebench.ai/table_2026_06_25.csv	224444814b13…	2026-07-14 00:15Z

Snapshots are immutable raw payloads — every published number traces back to one. Re-fetching an identical payload bumps the re-fetch time instead of storing a duplicate. Scores from this source appear on benchmark profiles and model pages with this source linked.