UnpredictaBench - Shuffling: leaderboard

Metric: KS@100 (%): share of the 19 Shuffling problems (random orderings of items) for which 100 samples from the model (100 independent calls at temperature 1.0, reasoning effort none, up to 5 retries for a parseable answer) are not rejected by a two-sample Kolmogorov-Smirnov test against ground-truth samples at p < 0.0001; higher is better. Source: arxiv.org. Saturation forecast: Around 2037. 20 models tracked.

Top models

#ModelScore
1Llama 3.2 1B Instruct36.84
2OLMo 3 7B Instruct36.84
3DeepSeek V3.2 (Non-reasoning)36.84
4Qwen 3.5 2B (Non-reasoning)36.84
5Nemotron 3 Nano 30B A3B (Non-reasoning)31.58
6Llama 3.1 70B Instruct21.05
7Llama 3.1 8B Instruct15.79
8GPT-4o5.26
9Ministral-3-3B-Instruct-25125.26
10Qwen 3.5 397B A17B (Non-reasoning)5.26
11Claude Sonnet 4.60
12GPT-4o Mini0
13Mistral Large 30
14Qwen 3 32B (Non-reasoning)0
15Phi-3.5-mini-instruct0

Interactive version: theaggregate.ai/benchmark?slug=unpredictabench-shuffling · How It Works · Data refreshed daily, snapshot 2026-09-29.