Laya Release - XNLI Other Languages: leaderboard

Laya T4 Colab run (research/results/t4_colab_benchmark.json): seed 13, 20 options per choice question, first 300 test rows per language; "Every checkpoint answered byte-identical questions" 300 per language (English: 300; other: 13 MASSIVE / 14 XNLI languages x 300) Metric: accuracy.

Metric: accuracy. Source: raw.githubusercontent.com. 1 model tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1Laya (421M)0.521

Interactive version: theaggregate.ai/benchmark?slug=laya-release-xnli-other-languages · How It Works · Data refreshed daily, snapshot 2026-11-16.