To method extended context prompts properly, versions need sturdy remember capabilities. The 'Needle In A Haystack' (NIAH) evaluation steps a product's capability to precisely remember details from the large corpus of knowledge. We Improved the robustness of the benchmark by utilizing one of thirty random needle/issue pairs for every prompt and tes