PredicateLongBench - Near-Sorted Decoys (LongBench): leaderboard
PredicateLongBench asks a model to find the run of words in a very long list that satisfies a stated predicate, such as a lexicographically sorted sequence. This board covers locating the sorted run in a LongBench v2 document seeded with eight near-sorted decoys scattered through the context. Answers are exact-match and outputs that exceed the generation budget count as wrong, so the score reflects both search and output discipline.
Metric: Accuracy (%; share of the 93 LongBench v2 documents under 170K tokens used as word lists whose answer is exactly the one word sequence satisfying the predicate, locate the sorted run among 8 same-length near-sorted decoys scattered through the context; up to 16K output tokens including reasoning, an answer past the cap counts as wrong). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Thinking) | 49 | #102 (Claude Opus 4.6) |
| 2 | GPT-5.4 (High) | 36 | #123 (GPT-5.4) |
| 3 | Gemini 3.1 Pro (Preview) (High) | 11 | #94 (Gemini 3.1 Pro (Preview)) |
| 4 | GLM-5.1 | 2 | #179 |
| 5 | GPT-5.4 (Non-reasoning) | 1 | #123 (GPT-5.4) |
| 6 | Claude Opus 4.6 | 0 | #102 |
| 7 | Qwen 3.5 397B A17B | 0 | #246 |
| 8 | MiniMax-M2.7 | 0 | #518 |
| 9 | Gemini 3.1 Pro (Preview) (Low) | 0 | #94 (Gemini 3.1 Pro (Preview)) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=predicatelongbench-near-sorted-decoys-longbench · How It Works · Data refreshed daily, snapshot 2026-11-13.