Laguna S 2.1 scored really low on AlmanBench, even lower than Ternary Bonsai 27B

1 · Onur Solmaz · July 23, 2026, 3:47 p.m.
X tweet Machine Learning benchmarking AI Models Performance Analysis
Summary
An analysis of the low performance of the Laguna S 2.1 model on AlmanBench compared to the Ternary Bonsai 27B, questioning the adequacy of the pretraining mix in terms of multilingual data.
Read full post on solmaz.io →
AUTHOR