BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

SimpleBench

Trick questions where humans score ~84%: spatio-temporal reasoning and social intelligence.

As of 24 Oct 2026, Claude Fable 5 leads SimpleBench on BenchLeader with 81.9%, ahead of Claude Opus 5 at 80.6%, across 87 model configurations with a published result.

Published by
SimpleBenchdata via Epoch AI Benchmarking Hub
Category
Reasoning
Index weight
1.0
Models
87
Data as of
24 Oct 2026
Newest result
12 Aug 2026

The publisher has posted no new results on this board since 12 Aug 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

What the test looks like

SimpleBench asks trick questions that ordinary people find easy and models find hard: spatial and temporal reasoning, social situations, and puzzles with an irrelevant detail meant to lure a memorised answer. A human baseline scores about 84%.

How it is scored

Percent of multiple-choice questions answered correctly, run by the benchmark's author with a fixed prompt.

What to keep in mind

It is small, so a few questions change the score, and the questions are private, which limits scrutiny of individual items. It is a good complement to knowledge tests precisely because it punishes pattern-matching.

Top 15
  1. 1Claude Fable 5 (max)81.9%
  2. 2Claude Opus 580.6%
  3. 3Gemini 3.1 Pro79.6%
  4. 4GPT-5.5 Pro (high)76.9%
  5. 5GPT-5.5 Pro76.9%
  6. 6Gemini 3.5 Flash76.7%
  7. 7Gemini 3 Pro76.4%
  8. 8Grok 4.675.9%
  9. 9Muse Spark 1.274.5%
  10. 10GPT-5.4 Pro74.1%
  11. 11GPT-5.6 Sol Pro (xhigh)71.7%
  12. 12GPT-5.6 Sol Pro71.7%
  13. 13Qwen3.7 (max)70.4%
  14. 14Grok 4.570.0%
  15. 15GPT-5.569.0%
87 of 87
#
1Claude Fable 5max81.9%69.32026-06-09
2Claude Opus 580.6%65.52026-07-24
3Gemini 3.1 Pro79.6%60.62026-02-19
4GPT-5.5 Prohigh76.9%–2026-04-23
5GPT-5.5 Pro76.9%–2026-04-23
6Gemini 3.5 Flash76.7%49.12026-05-19
7Gemini 3 Pro76.4%60.42025-11-18
8Grok 4.675.9%57.82026-08-12
9Muse Spark 1.274.5%–2026-08-05
10GPT-5.4 Pro74.1%63.32026-03-05
11GPT-5.6 Sol Proxhigh71.7%–2026-07-09
12GPT-5.6 Sol Pro71.7%–2026-07-09
13Qwen3.7max70.4%60.62026-05-19
14Grok 4.570.0%59.02026-07-08
15GPT-5.569.0%60.32026-04-23
16Claude Opus 4.667.6%62.62026-02-05
17GPT-5.6 Solxhigh64.8%66.62026-07-09
18Claude Opus 4.864.8%60.12026-05-28
19GPT-5.6 Sol64.8%55.92026-07-09
20Qwen3.6 Maxmax63.0%62.32026-04-20
21Gemini 2.5 Pro62.4%53.92025-06-05
22Claude Opus 4.562.0%57.92025-11-24
23Claude Opus 4.761.7%63.32026-04-16
24GPT-5 Pro61.6%59.82025-10-07
25GPT-5 Prohigh61.6%–2025-10-07
26Gemini 3 Flash61.1%57.02025-12-17
27DeepSeek V4 Flash61.1%52.22026-07-31
28Kimi K3max60.7%64.72026-07-16
29Claude Sonnet 560.6%56.12026-06-30
30Grok 460.5%58.02025-07-09
31Claude Opus 4.160.0%51.52025-08-05
32Claude Opus 458.8%52.62025-05-22
33GLM 5.258.8%52.02026-06-16
34Kimi K2.7 Code57.9%55.82026-06-12
35GPT-5.2 Pro57.4%–2025-12-11
36GPT-5high56.7%57.92025-08-07
37Grok 4.1 Fastthinking56.0%54.72025-11-19
38Grok 4.1 Fastno reasoning56.0%42.52025-11-19
39GLM 5.155.1%55.92026-04-07
40Claude Sonnet 4.554.3%53.72025-09-29
41GPT-5.1high53.2%57.72025-11-13
42GLM-553.2%54.92026-02-11
43o3high53.1%53.92025-04-16
44DeepSeek V3.2 Speciale52.6%53.02025-12-01
45GPT-5.6 Terraxhigh48.9%62.82026-07-09
46GPT-5.6 Terra48.9%52.72026-07-09
47GLM 4.747.7%51.92025-12-22
48GPT-5.6 Lunaxhigh46.8%59.42026-07-09
49Kimi K2.546.8%54.52026-01-27
50GPT-5.6 Luna46.8%48.02026-07-09
51Claude 3.7 Sonnet46.4%50.02025-02-24
52GPT-5.245.8%58.02025-12-11
53MiniMax M345.8%56.42026-06-01
54GPT-5.2high45.8%56.02025-12-11
55Claude Sonnet 445.5%50.52025-05-22
56o141.7%53.12024-09-12
57Claude 3.5 Sonnet41.4%46.32024-10-22
58Gemini 2.5 Flash41.2%52.72025-06-17
59DeepSeek R1 052840.8%49.52025-05-28
60o1high40.1%48.32024-12-17

More reasoning benchmarks · All LLM benchmarks

Cite as: BenchLeader, “SimpleBench leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/simplebench.

SimpleBench: questions

What does SimpleBench measure?
SimpleBench asks trick questions that ordinary people find easy and models find hard: spatial and temporal reasoning, social situations, and puzzles with an irrelevant detail meant to lure a memorised answer. A human baseline scores about 84%. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads SimpleBench?
Claude Fable 5 leads SimpleBench with 81.9% as of 24 Oct 2026, ahead of Claude Opus 5 at 80.6%.
How many models have SimpleBench results?
87 model configurations have a SimpleBench result on BenchLeader, all taken from SimpleBench via Epoch AI Benchmarking Hub.
Who runs SimpleBench and how often is it updated?
SimpleBench is published by SimpleBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does SimpleBench count toward the BenchLeader Index?
Yes. SimpleBench contributes to the reasoning category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.