Poolside Launches Laguna XS.2 (Apache 2.0, 68.2% SWE-bench) and M.1 (72.5%) Agentic Coding Models

Poolside AI·high signal

Poolside AI released two Laguna family models: M.1 (225B total / 23B active MoE) scoring 72.5% on SWE-bench Verified, and XS.2 (33B total / 3B active) scoring 68.2% — released under Apache 2.0 and runnable locally on a Mac with 36GB RAM via Ollama. Alongside the models, Poolside shipped 'pool', a lightweight terminal-based coding agent and ACP client-server, the same environment used internally for agent RL training. XS.2 also scores 44.5% on SWE-bench Pro and 30.1% on Terminal-Bench 2.0.