Laguna XS.2 and M.1
The company has released two new models in the Laguna family, Laguna M.1 and Laguna XS.2, designed for agentic coding and long-horizon tasks. Laguna M.1 is a 225B-parameter Mixture of Experts model trained in-house, while Laguna XS.2 is a smaller, open-weight model with 33B total parameters. Both models are available for free use via API and OpenRouter, with XS.2's weights released under an Apache 2.0 license.
- ▪Laguna M.1 is a 225B-parameter Mixture of Experts model with 23B activated parameters, trained on 30T tokens using 6,144 NVIDIA Hopper GPUs.
- ▪Laguna XS.2 is a 33B-parameter MoE model with 3B activated parameters and is the company's first open-weight release under the Apache 2.0 license.
- ▪Laguna M.1 achieved 46.9% on SWE-bench Pro and 40.7% on Terminal-Bench 2.0, while XS.2 reached 44.5% and 30.1% respectively.
- ▪Both models are built for agentic coding and long-horizon tasks, emphasizing code execution as a more expressive interface than tool calling.
- ▪The models were developed by the company's 60-person Applied Research team and are available in preview through API and OpenRouter.
Opening excerpt (first ~120 words) tap to expand
2026-04-28 Laguna XS.2 and M.1: A Deeper Dive Poolside team Table of contents We’ve released the first two models in the Laguna family, Laguna M.1 and Laguna XS.2, alongside the runtime we use to train and operate agents, available through two product experiences in preview.Open weightsComing soonWorking with NVIDIAModel buildingData and automixingMuonAgent RLGet started We’ve released the first two models in the Laguna family, Laguna M.1 and Laguna XS.2, alongside the runtime we use to train and operate agents, available through two product experiences in preview.Laguna M.1 came first, finishing pre-training at the end of last year; it's the foundation for everything else we're building across the family.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Poolside.
Discussion
0 commentsMore from Poolside
