Poolside

Laguna M.1Poolside

Laguna M.1 is Poolside's flagship agentic coding model — a 225B total parameter MoE with 23B active parameters per token, trained from scratch on 30T tokens. Optimized for multi-step software engineering tasks: codebase exploration, file editing, running tests, and iterating on fixes. Reaches 72.5% on SWE-Bench Verified and 65.4% on SWE-Bench Multilingual. Best used in agentic workflows rather than as a standalone chat model.

Laguna S 2.1Poolside

Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model from Poolside with only 8B active parameters per token, built for agentic coding and long-horizon reasoning. It scores 70.2% on Terminal-Bench 2.1 and 78.5% on SWE-Bench Multilingual — leading all open disclosed-size models in its class. Supports thinking and no-thinking modes, tool calling, vision, and a 1M token context window.