UnpredictaBench - Text: leaderboard
KS@100 on the 160 Text problems, which name or describe a canonical distribution to sample from, from UnpredictaBench (2026), a benchmark of sampling from target distributions: the percent of problems for which 100 samples from the model, drawn at temperature 1.0 with reasoning off, are not rejected by a two-sample Kolmogorov-Smirnov test against ground-truth samples at p below 0.0001. Higher is better.
Metric: KS@100 (%): share of the 160 Text problems (named or described canonical distributions) 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: Not forecast. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 (Non-reasoning) | 28.13 | #406 (DeepSeek V3.2) |
| 2 | GPT-5.4 (Non-reasoning) | 25.63 | #123 (GPT-5.4) |
| 3 | Llama 3.1 70B Instruct | 23.13 | #1546 |
| 4 | GPT-4o | 21.25 | #741 |
| 5 | Ministral-3-3B-Instruct-2512 | 21.25 | #2579 |
| 6 | nemotron-3-nano-30B-a3B (Non-reasoning) | 20.63 | #1875 (nemotron-3-nano-30B-a3B) |
| 7 | Qwen 3.5 2B (Non-reasoning) | 14.38 | #4810 (Qwen 3.5 2B) |
| 8 | Qwen 3 32B (Non-reasoning) | 11.88 | #960 (Qwen 3 32B) |
| 9 | Llama 3.2 1B Instruct | 10 | #9047 |
| 10 | GPT-4o Mini | 9.38 | #1623 |
| 11 | OLMo 3 7B Instruct | 7.5 | #6860 |
| 12 | Grok 4.1 Fast (Non-reasoning) | 6.88 | #430 (Grok 4.1 Fast) |
| 13 | Claude Sonnet 4.6 | 5.03 | #161 |
| 14 | Qwen 3.5 397B A17B (Non-reasoning) | 3.8 | #246 (Qwen 3.5 397B A17B) |
| 15 | Llama 3.1 8B Instruct | 3.75 | #4246 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=unpredictabench-text · How It Works · Data refreshed daily, snapshot 2026-11-13.