---
title: "Laguna S 2.1 scored really low on AlmanBench, even lower than Ternary Bonsai 27B"
date: 2026-07-23
canonical: https://solmaz.io/x/2080276261050741080/
x_url: https://x.com/onusoz/status/2080276261050741080
license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
---

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

A 118B, 8B active parameter model, scored lower in FB8 precision, than a ternary quantized 27B model 🤨

Did the pretraining mix lack multilingual data by a lot? 🤔

https://alman.ai/almanbench