SimpleBench
Integrity rank #17 of 61 · 71 models scored · top score 78.28 · Claude Fable 5
Basic human common-sense reasoning that resists memorized knowledge
Strongest on discrimination, weakest on freshness.
Reported scores — protocols may differ. How we compare
| # | Model | Score | Evidence | Measured |
|---|---|---|---|---|
| 1 | Claude Fable 5 | 78.28 | unverifiedT1 | 2026-06-09 |
| 2 | Gemini 3.1 Pro | 75.52 | unverifiedT1 | 2026-02-19 |
| 3 | GPT-5.5 Pro | 72.28 | unverifiedT1 | 2026-04-23 |
| 4 | Gemini 3.5 Flash | 72.04 | unverifiedT1 | 2026-05-19 |
| 5 | Gemini 3 Pro | 71.68 | unverifiedT1 | 2025-11-18 |
| 6 | GPT-5.4 Pro | 68.92 | unverifiedT1 | 2026-03-05 |
| 7 | GPT-5.6 Sol | 66.04 | unverifiedT1 | 2026-07-09 |
| 8 | Qwen3.7-Max | 64.48 | unverifiedT1 | 2026-05-19 |
| 9 | Grok 4.5 | 64 | unverifiedT1 | 2026-07-08 |
| 10 | GPT-5.5 | 62.8 | unverifiedT1 | 2026-04-23 |
| 11 | Claude Opus 4.6 | 61.12 | unverifiedT1 | 2026-02-05 |
| 12 | Claude Opus 4.8 | 57.76 | unverifiedT1 | 2026-05-28 |
| 13 | Qwen 3.6 Max (Preview) | 55.6 | unverifiedT1 | 2026-04-20 |
| 14 | Claude Opus 4.7 | 55.48 | unverifiedT1 | 2026-04-16 |
| 15 | Gemini 2.5 Pro (Jun 2025) | 54.88 | unverifiedT1 | 2025-06-05 |
| 16 | Claude Opus 4.5 | 54.4 | unverifiedT1 | 2025-11-24 |
| 17 | GPT-5 Pro | 53.92 | unverifiedT1 | 2025-10-07 |
| 18 | DeepSeek-V4-Pro | 53.44 | unverifiedT1 | 2026-04-24 |
| 19 | Gemini 3 Flash | 53.32 | unverifiedT1 | 2025-12-17 |
| 20 | Claude Sonnet 5 | 52.72 | unverifiedT1 | 2026-06-30 |
| 21 | Grok 4 | 52.6 | unverifiedT1 | 2025-07-09 |
| 22 | Claude Opus 4.1 | 52 | unverifiedT1 | 2025-08-05 |
| 23 | Claude Opus 4 | 50.56 | unverifiedT1 | 2025-05-22 |
| 24 | Kimi K2.7 Code | 49.48 | unverifiedT1 | 2026-06-12 |
| 25 | GPT-5.2 Pro | 48.88 | unverifiedT1 | 2025-12-11 |
| 26 | GPT-5 | 48.04 | unverifiedT1 | 2025-08-07 |
| 27 | GLM-5.1 | 46.12 | unverifiedT1 | 2026-04-07 |
| 28 | Claude Sonnet 4.5 | 45.16 | unverifiedT1 | 2025-09-29 |
| 29 | GLM-5 | 43.84 | unverifiedT1 | 2026-02-11 |
| 30 | GPT-5.1 | 43.84 | unverifiedT1 | 2025-11-13 |
| 31 | o3 | 43.72 | unverifiedT1 | 2024-12-20 |
| 32 | Gemini 2.5 Pro (Mar 2025) | 41.92 | unverifiedT1 | 2025-03-25 |
| 33 | GPT-5.6 Terra | 38.68 | unverifiedT1 | 2026-07-09 |
| 34 | GLM-4.7 | 37.24 | unverifiedT1 | 2025-12-22 |
| 35 | Kimi K2.5 | 36.16 | unverifiedT1 | 2026-02-02 |
| 36 | GPT-5.6 Luna | 36.16 | unverifiedT1 | 2026-07-09 |
| 37 | Claude 3.7 Sonnet | 35.68 | unverifiedT1 | 2025-02-24 |
| 38 | MiniMax-M3 | 34.96 | unverifiedT1 | 2026-06-01 |
| 39 | GPT-5.2 | 34.96 | unverifiedT1 | 2025-12-11 |
| 40 | Claude Sonnet 4 | 34.6 | unverifiedT1 | 2025-05-22 |
| 41 | o1-preview | 30.04 | unverifiedT1 | 2024-09-12 |
| 42 | Claude 3.5 Sonnet (October 2024) | 29.68 | unverifiedT1 | 2024-10-22 |
| 43 | Gemini 2.5 Flash (Jun 2025) | 29.44 | unverifiedT1 | 2025-06-17 |
| 44 | DeepSeek-R1 (May 2025) | 28.96 | unverifiedT1 | 2025-05-28 |
| 45 | o1 | 28.12 | unverifiedT1 | 2024-12-05 |
| 46 | o4-mini | 26.44 | unverifiedT1 | 2025-04-16 |
| 47 | Grok 3 | 23.32 | unverifiedT1 | 2025-02-17 |
| 48 | Qwen 3.6 Flash | 22.24 | unverifiedT1 | 2026-04-26 |
| 49 | GPT-4.5 | 21.4 | unverifiedT1 | 2025-02-27 |
| 50 | Gemini 2.0 Flash (Dec 2024) | 17.32 | unverifiedT1 | 2024-12-11 |
| 51 | Qwen3-235B-A22B | 17.2 | unverifiedT1 | 2025-04-28 |
| 52 | DeepSeek-R1 | 17.08 | unverifiedT1 | 2025-01-20 |
| 53 | Gemini 2.0 Flash Thinking (Jan 2025) | 16.84 | unverifiedT1 | 2025-01-21 |
| 54 | Llama 4 Maverick | 13.24 | unverifiedT1 | 2025-04-05 |
| 55 | Claude 3.5 Sonnet | 13 | unverifiedT1 | 2024-06-20 |
| 56 | DeepSeek-V3 (Mar 2025) | 12.64 | unverifiedT1 | 2025-03-24 |
| 57 | Gemini 1.5 Pro (Sept 2024) | 12.52 | unverifiedT1 | 2024-09-24 |
| 58 | GPT-4.1 | 12.4 | unverifiedT1 | 2025-04-14 |
| 59 | Kimi K2 (Jul 2025) | 11.56 | unverifiedT1 | 2025-07-11 |
| 60 | GPT-4 Turbo (Apr 2024) | 10.12 | unverifiedT1 | 2024-04-09 |
| 61 | Claude 3 Opus | 8.2 | unverifiedT1 | 2024-03-04 |
| 62 | Llama 3.1-405B | 7.6 | unverifiedT1 | 2024-07-23 |
| 63 | o3-mini | 7.36 | unverifiedT1 | 2025-01-31 |
| 64 | Grok-2 (Dec 2024) | 7.24 | unverifiedT1 | 2024-08-13 |
| 65 | Mistral Large 2 (Jul 2024) | 7 | unverifiedT1 | 2024-07-24 |
| 66 | gpt-oss-120b | 6.52 | unverifiedT1 | 2025-08-05 |
| 67 | Llama 3.3 70B | 3.88 | unverifiedT1 | 2024-12-06 |
| 68 | DeepSeek-V3 | 2.68 | unverifiedT1 | 2024-12-24 |
| 69 | o1-mini | 1.72 | unverifiedT1 | 2024-09-12 |
| 70 | GPT-4o (Aug 2024) | 1.36 | unverifiedT1 | 2024-05-13 |
| 71 | GPT-4o mini | 0 | unverifiedT1 | 2024-07-18 |
Continue from this leaderboard into the tightest head-to-head reads.
Weighted composite. Each dimension opens to the source-backed sub-signals it is computed from.
Claims drawn from cited facts, not live model generation.
Between the 71-model sample and the 78.28-point range, this table has both the depth and the separation to support reading its ranks closely. This wide spread supports real separation between models, meaning rank differences are trustworthy.
3 cited facts
The benchmark's Integrity Index of 98 (grade A) ranks it 13th out of 53 benchmarks; its weakest component is age, which means the benchmark may not discriminate the newest models as effectively.
4 cited facts
This benchmark is not saturated, which indicates that there remains genuine room to separate the field at the top. Just under seventy-eight points, with over twenty in reserve and a single model holding the top — the separation is real for now, but it rests on that model's result alone and the field has room to answer.
4 cited facts
Scores from this harness are directly comparable because the harness is consistent across evaluations, and the test set is held-out, meaning it is not public, which reduces the risk of contamination. No confirmed train-set leakage has been found; the held-out design serves as the primary defense, and harness sensitivity (e.g., fixed system prompt and answer extraction) means third-party scores depend on maintainers running the private set.
4 cited facts
Hand-written adversarial questions are the defense against memorized knowledge. The design's bet is that failure here reflects a genuine common-sense gap rather than a missing fact, and that bet is what to test the scores against. Only a handful of questions are released publicly; the bulk remain private, limiting independent scrutiny. It assumes these questions are easy for average humans but hard for models, isolating reasoning from memorization, and that the held-out set generalizes as a reasoning proxy. The sharpest caveat is that the human baseline rests on an extremely small sample, and the small number of items amplifies per-question noise, so results may not reflect true reasoning ability.
5 cited facts
SimpleBench: A multiple-choice text benchmark of adversarial everyday-reasoning questions (spatio-temporal, social, and trick questions) on which unspecialized humans outperform frontier models. 213 questions total but only 10 are public; the rest are held out.
Claude Fable 5 leads SimpleBench at 78.28. The full leaderboard above lists every recorded measurement, not just the headline number.
71 models have recorded scores on SimpleBench, spanning a score spread of 78.28.
tensor.news grades SimpleBench A for integrity (score 98/100), ranking #17 of 61 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — SimpleBench still has headroom and continues to discriminate between models rather than bunching them at the ceiling.
Its test set is held-out, which lowers — but does not eliminate — the risk that training data leaked into the questions.
Scores are reported under a consistent harness, so comparisons on SimpleBench are reasonably apples-to-apples.
Every SimpleBench measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites simple-bench.com.