Poolside AI veröffentlicht erste öffentliche Modelle: Laguna M.1 & XS.2

Poolside AI Ships First Public Models: Laguna M.1 & XS.2

Poolside AI has released its first publicly available models, Laguna M.1 and Laguna XS.2. The XS.2 is a 33B-total / 3B-active MoE model purpose-built for agentic coding and long-horizon tasks, capable of running on a single GPU. Both models were trained from scratch on Poolside's own data pipelines, training infrastructure, and agent reinforcement learning stack. Laguna XS.2 ranks #12 on SWE-Bench Pro, close to Qwen3.6, and is released under Apache 2.0.

Why It Matters

Poolside's full-stack ownership thesis — control the data, training, RL, and inference — now has a public proof point. The Apache 2.0 release makes Laguna XS.2 immediately deployable for enterprise coding agent workflows without licensing friction.