UnpredictaBench - Code: leaderboard

Metric: KS@100 (%): share of the 238 Code problems (distributions induced by stochastic programs) 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 2040. 20 models tracked.

Top models

#ModelScore
1GPT-4o29.41
2Nemotron 3 Nano 30B A3B (Non-reasoning)18.49
3Ministral-3-3B-Instruct-251217.23
4DeepSeek V3.2 (Non-reasoning)14.29
5Qwen 3.5 2B (Non-reasoning)12.61
6GPT-4o Mini10.5
7GPT-5.4 (Non-reasoning)10.5
8Llama 3.1 70B Instruct9.66
9OLMo 3 7B Instruct5.46
10Qwen 3 32B (Non-reasoning)5.46
11Grok 4.1 Fast (Non-reasoning)5.46
12Claude Sonnet 4.65.04
13Mistral Large 35.04
14Llama 3.1 8B Instruct4.62
15Llama 3.2 1B Instruct4.2

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