NoLiMa 64K

reasoning

NoLiMa evaluated at a 65536-token context length. Tests latent associative reasoning in long contexts with minimal lexical overlap between questions and needles.

Leaderboard