PredicateLongBench - Lexicographic Locate (Synthetic): 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, locate the one lexicographically sorted run of k=5 words, no decoys; 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
1Claude Opus 4.6 (Thinking)97
2GPT-5.4 (High)62
3Gemini 3.1 Pro (Preview) (Low)45
4Gemini 3.1 Pro (Preview) (High)43
5Claude Opus 4.625
6GLM-5.18
7MiniMax-M2.78
8GPT-5.4 (Non-reasoning)2
9Qwen 3.5 397B A17B0

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