NoLiMa 128K

reasoning

NoLiMa evaluated at a 131072-token context length. Tests latent associative reasoning in long contexts with minimal lexical overlap between questions and needles.

Leaderboard