Can LLMs reason over extended multilingual contexts? Towards long-context evaluation beyond retrieval and haystacks

Amey Hengle, Prasoon Bajpai, Soham Dan +1

Existing multilingual long-context benchmarks, often based on the popular needle-in-a-haystack test, primarily evaluate a model's ability to locate specific information buried with…