Topic

Laguna S 2 1

Coverage tagged laguna s 2 1.

Articles: 1
Desk: 1

Data graphic: Poolside's Laguna S 2.1, a 118B open-weight MoE that uses about 8B parameters per token, scores 70.2 on Terminal-Bench 2.1. That was 11th on Poolside's launch board but is 25th of 39 on the public board as of 13 Sep 2026. It trails DeepSeek-V4.1-Flash 90.6 by 20.4 points at the same ~8B active parameters, and also trails GLM-5.3-Flash 84.3 and Qwen3.8-27B 73.0 . A strip shows its 256 + 1 experts, ~72 GB of 4-bit weights and 19.5 tok/s single-stream on DGX Spark.

stableResearch Laguna S 2 1

Laguna S 2.1 Coding Benchmark Performance: How Efficient Open Models Are Reshaping Enterprise AI

The open-weight MoE model targets teams squeezed between frontier API costs, inference latency and data-residency rules that rule out hosted inference.

Jul 22 · 14 min