Atlas · C07 · Core cognition
Knowledge
Stored facts and understanding across fields (CHC: comprehension-knowledge).
AI Reliable score
8 / 10
Expert level · peak 11 · trend ↑ · grade A
Why this score
GPQA Diamond: top model 96% vs PhD domain experts 65%. Breadth across all fields at once is beyond any human (Peak 11). Reliable held at 8 because confident factual errors still occur and frontier-hard questions (Humanity's Last Exam) top out at ~55-61%.
On the human scale
2Beginner. Basic general knowledge
4Average adult. Average educated adult
6Competent pro. Professional: solid knowledge of own field
8Expert. Expert: PhD / specialist depth ← AI today
10Best humans. Leading authorities
Future warning
At its best, AI already reaches 11 on this capability. Peak results are never counted in the index; they show where the water may rise next.
Evidence
GPQA Diamond (Artificial Analysis snapshot via modelgrep) + original GPQA paperA
E09 · 2026-11 / 2023-11
- Measures
- Graduate-level 'Google-proof' science questions
- AI result
- Top: GPT-6 Astra 96.1%.
- Humans
- PhD domain experts 65% (74% discounting clear mistakes); skilled non-experts with web 34%
- On our scale
- Above PhD experts on this test: expert to beyond-expert.
modelgrep.com ↗
arxiv.org ↗
Humanity’s Last Exam (two leaderboards)B
E10 · 2026-09 to 2026-11
- Measures
- Expert-written questions at the frontier of human knowledge
- AI result
- Top 54.8% (benchleader) to 61.4% (Artificial Analysis): evaluators differ.
- Humans
- Not published
- On our scale
- Frontier questions still unsolved ~40-45% of the time: supports Reliable 8, not 10.
www.benchleader.com ↗
modelgrep.com ↗
A Definition of AGI (Hendrycks et al., CHC-based)B
E07 · 2025-10
- Measures
- 10 CHC cognitive domains, each worth 10% of an 'AGI score'
- AI result
- GPT-5 total 57%: Knowledge 9, Reading-Writing 10, Maths 10, Reasoning 7, Working memory 4, LT memory storage 0, LT memory retrieval 4, Visual 4, Auditory 6, Speed 3 (out of 10 each).
- Humans
- Well-educated adult (definition of 100%)
- On our scale
- Used as a cross-check on the profile shape, not for absolute levels (single model, dated Oct 2025).
arxiv.org ↗
Vectara hallucination leaderboard (next generation)B
E11 · 2025-11
- Measures
- Share of document summaries containing unsupported claims
- AI result
- Best 3.3%; several frontier models above 10% (e.g. Gemini-3-pro 13.6%).
- Humans
- Not published
- On our scale
- Confident errors remain common: limits self-monitoring.
vectara.com ↗
History
First reading: Atlas v0.1, 5 Nov 2026. Every later change will be listed here with its reason.
Disagree? Send a correction
