BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

AA-Omniscience

Knowledge recall that penalises hallucination: correct answers minus confident wrong ones, from −100 to 100. Run by Artificial Analysis.

As of 25 Oct 2026, Claude Opus 5.5 leads AA-Omniscience on BenchLeader with 46.4, ahead of GPT-6 Astra at 43.7, across 552 model configurations with a published result.

Published by
Artificial Analysis
Category
Knowledge
Index weight
1.0
Models
552
Data as of
25 Oct 2026
Newest result
23 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

What the test looks like

AA-Omniscience asks thousands of factual questions across many domains and rewards knowing when not to answer: a correct answer scores positively, a wrong answer negatively, and an abstention zero.

How it is scored

A score from -100 to 100, correct minus confidently wrong, run by Artificial Analysis. Most models score below zero.

What to keep in mind

The penalty design means a cautious model can outscore a more knowledgeable but over-confident one. Read it with the hallucination rate that Artificial Analysis publishes alongside it.

Top 15
  1. 1Claude Opus 5.5 (max)46.4
  2. 2GPT-6 Astra (high)43.7
  3. 3Claude Fable 5.1 (max)43.5
  4. 4GPT-6 Astra (xhigh)43.4
  5. 5GPT-6 Astra (max)43.4
  6. 6Claude Fable 5 (max)43.3
  7. 7Claude Opus 5.5 (xhigh)42.6
  8. 8Claude Fable 5.1 (xhigh)42.4
  9. 9Gemini 4 Argon (high)42.4
  10. 10GPT-6 Astra (medium)42.2
  11. 11GPT-6.1 Sol (max)41.5
  12. 12GPT-6.1 Sol (high)41.5
  13. 13GPT-6.1 Sol (xhigh)40.9
  14. 14Claude Fable 5.1 (high)40.8
  15. 15Claude Opus 5.5 (high)40.6
552 of 552
#
1Claude Opus 5.5max46.472.62026-09-22
2GPT-6 Astrahigh43.769.62026-09-03
3Claude Fable 5.1max43.570.52026-09-01
4GPT-6 Astraxhigh43.468.92026-09-03
5GPT-6 Astramax43.470.42026-09-03
6Claude Fable 5max43.369.32026-06-09
7Claude Opus 5.5xhigh42.670.02026-09-22
8Claude Fable 5.1xhigh42.469.52026-09-01
9Gemini 4 Argonhigh42.470.72026-09-30
10GPT-6 Astramedium42.267.52026-09-03
11GPT-6.1 Solmax41.569.92026-09-29
12GPT-6.1 Solhigh41.567.72026-09-29
13GPT-6.1 Solxhigh40.967.72026-09-29
14Claude Fable 5.1high40.869.42026-09-01
15Claude Opus 5.5high40.669.42026-09-22
16GPT-6 Astralow40.565.92026-09-03
17Claude Opus 5.5medium40.366.52026-09-22
18GPT-6.1 Solmedium40.067.22026-09-29
19Claude Opus 5.5low38.964.72026-09-22
20GPT-6.1 Sollow37.664.92026-09-29
21Claude Fable 5.1medium37.667.32026-09-01
22Claude Opus 5max37.168.82026-07-24
23Claude Opus 5xhigh35.467.42026-07-24
24Claude Fable 5.1low34.165.02026-09-01
25Claude Opus 5high33.767.92026-07-24
26Claude Sonnet 5.5max32.369.42026-09-28
27Grok 4.7xhigh32.061.62026-09-21
28Gemini 3.1 Pro31.960.52026-02-19
29Claude Opus 5medium31.064.62026-07-24
30Grok 4.7high30.963.92026-09-21
31Grok 4.6high30.562.92026-08-12
32Muse Wonder Fullmax30.170.62026-10-12
33Grok 4.7low29.661.02026-09-21
34Gemini 3.8 Flashhigh29.664.42026-09-02
35NewKimi K3.1 Pro29.566.92026-10-23
36Grok 4.6xhigh29.362.52026-08-12
37Claude Opus 4.8max28.863.32026-05-28
38Gemini 3.8 Flashmedium28.664.42026-09-02
39Claude Opus 5low28.661.32026-07-24
40Muse Spark 1.1xhigh28.161.62026-07-09
41Grok 4.6medium2864.22026-08-12
42Claude Opus 4.7max27.362.42026-04-16
43Muse Spark 1.2xhigh27.263.22026-08-05
44GPT-6 Solmax27.166.42026-09-22
45GPT-6 Solmedium27.062.62026-09-22
46GPT-6 Solhigh26.863.72026-09-22
47GPT-6 Solxhigh26.764.32026-09-22
48GPT-6 Sollow26.559.72026-09-22
49Gemini 3.7 Flashhigh26.563.52026-08-13
50Grok 4.6low25.959.32026-08-12
51Muse Wondermax25.567.52026-10-12
52Grok 4.5high25.359.32026-07-08
53Muse Spark 1.3max2567.12026-09-02
54Muse Spark Openxhigh24.662.02026-10-15
55Gemini 3.7 Flashmedium23.763.12026-08-13
56Claude Sonnet 5.5xhigh23.567.92026-09-28
57Muse Spark 1.3xhigh23.164.92026-09-02
58Gemini 3.6 Flashhigh22.162.02026-07-21
59Gemini 3.7 Flashlow22.160.32026-08-13
60GPT-5.6 Solmax22.068.02026-07-09

More knowledge benchmarks · All LLM benchmarks

Cite as: BenchLeader, “AA-Omniscience leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/aa_omniscience.

AA-Omniscience: questions

What does AA-Omniscience measure?
AA-Omniscience asks thousands of factual questions across many domains and rewards knowing when not to answer: a correct answer scores positively, a wrong answer negatively, and an abstention zero. Scores are reported in score; higher is better.
Which AI model leads AA-Omniscience?
Claude Opus 5.5 leads AA-Omniscience with 46.4 as of 25 Oct 2026, ahead of GPT-6 Astra at 43.7.
How many models have AA-Omniscience results?
552 model configurations have a AA-Omniscience result on BenchLeader, all taken from Artificial Analysis.
Who runs AA-Omniscience and how often is it updated?
AA-Omniscience is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does AA-Omniscience count toward the BenchLeader Index?
Yes. AA-Omniscience contributes to the knowledge category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.