BenchAtlas

Sources / vals-ai

Vals AI

healthydefault: independent

www.vals.ai · trust rank 70 · adapter vals-ai

Observations
1,036
append-only score facts
Raw snapshots
11
immutable fetched payloads
Last success
2026-07-14 00:16Z
end of last succeeded/partial run
Freshness
1.2 h
hours since last success

License & attribution

No ToS page or license statement exists on vals.ai; attributed as "Vals AI (vals.ai)" with a link as good practice.

What we ingest

Leaderboard JSON from per-benchmark pages: accuracy, stderr, cost and full run configs across ~30 benchmarks including their own professional-domain (legal, finance, medical) evals, which land as independent. The Vals Index is stored as their own family, never merged with ours.

Ingestion runs

last 3 of the run history
StartedDurationStatus
2026-07-14 00:16Z5.6 ssucceeded
2026-07-14 00:09Z5.7 ssucceeded
2026-07-13 20:03Z7.1 spartial

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 5
FetchedURLSHA-256
2026-07-13 20:03Zhttps://www.vals.ai/benchmarks/vals_indexed43197ac520
2026-07-13 20:03Zhttps://www.vals.ai/benchmarks/mmmufbab2c3e2f2b
2026-07-13 20:03Zhttps://www.vals.ai/benchmarks/mmlu_pro4a4e699eba45
2026-07-13 20:03Zhttps://www.vals.ai/benchmarks/gpqa0c51d101492d
2026-07-13 20:03Zhttps://www.vals.ai/benchmarks/math500e26048cb92a3

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.