PredicateLongBench - Unary Exists: leaderboard

Metric: Accuracy (%; share of the 100 synthetic word lists of about 128K tokens whose answer is exactly the one word sequence satisfying the predicate, any contiguous run of words matching a prefix, suffix or string-match predicate, no distractors; up to 16K output tokens including reasoning, an answer past the cap counts as wrong). Source: arxiv.org. Saturation forecast: Around December 2026. 9 models tracked.

Top models

#ModelScore
1GPT-5.4 (Non-reasoning)96
2GPT-5.4 (High)95
3Gemini 3.1 Pro (Preview) (High)93
4Gemini 3.1 Pro (Preview) (Low)91
5Claude Opus 4.6 (Thinking)87
6GLM-5.176
7Claude Opus 4.653
8MiniMax-M2.750
9Qwen 3.5 397B A17B45

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