Atlas

Benchmarks

← All benchmarks

FACTS Benchmark Suite (Grounding v3, Parametric v2)

General QA

The FACTS Benchmark Suite average after Kaggle moved two of its four component boards to new versions in October 2026: the unweighted mean of FACTS Grounding v3, FACTS Multimodal v1, FACTS Parametric v2 and FACTS Search v2. Only rows that report all four of these component versions are imported; the leaderboard's Average for a row missing a component is a mean over fewer boards and is not stored. Grounding and Parametric differ from the earlier suite's Grounding v2 and Parametric v1, so these scores are a separate scale from the FACTS Benchmark Suite and its incomplete-cohort diagnostics.

Top models (higher is better)

ModelScore
GPT-6 Astra75.4
GPT-6.1 Sol74.6
Gemini 3.8 Flash68.5
Claude Opus 5.564.0
GPT-6 Luna62.3
GPT-5.6 Terra50.0
Loading Atlas data…