NoLiMa 64K

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

id nolima-64k · max 1 · 1 models reported