AA-LCR

AA-LCR (Long Context Retrieval) tests whether a model actually retrieves information from very long documents. Relevant passages are hidden across documents of hundreds of thousands of tokens and the model must reliably find and combine them.

AA-LCR measures reading and reasoning over very long inputs. In detail, it checks:

• retrieval accuracy in documents of extreme length
• several hidden 'needles' that have to be found simultaneously
• resistance to being distracted by irrelevant filler material
• robustness close to the model's context window limit

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the AA-LCR benchmark?

AA-LCR (Long Context Retrieval) tests whether a model actually retrieves information from very long documents. Relevant passages are hidden across documents of hundreds of thousands of tokens and the model must reliably find and combine them.

What does AA-LCR test?

AA-LCR mainly measures reading and reasoning over very long inputs. Concretely, it checks: retrieval accuracy in documents of extreme length; several hidden 'needles' that have to be found simultaneously; resistance to being distracted by irrelevant filler material.

How is the AA-LCR score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get AA-LCR data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/aa-lcr/scores) or pull it directly from the Open LLM Radar API with any HTTP client.