Dataset · ArtificialAnalysis
AA-LCR Long-Context Logic Benchmark
ArtificialAnalysis/AA-LCR
100 complex questions over large document sets (~100k tokens) to test deep text analysis.
What's inside
The dataset includes 100 complex text and table questions requiring reasoning across multiple real-world documents simultaneously. Each document set averages around 100,000 input tokens, and answers cannot be found via direct search—requiring the synthesis of information from different sources. Version 1.1 from September 2026 re-verified and corrected 16 ground-truth answers.
What you can do with it
- Testing LLMs with large context windows for genuine text comprehension rather than simple search
- Evaluating RAG systems when working with numerous bulky documents
- Size
- up to 1K rows
- Downloads this month
- 24.4K
- Likes
- 40
How to load it
from datasets import load_dataset
ds = load_dataset("ArtificialAnalysis/AA-LCR")