Poolside AI Releases Laguna XS.2 and M.1 Open-Weight Coding Models

Poolside AI released two new open-weight agentic coding models, Laguna XS.2 and Laguna M.1, designed for long-horizon coding tasks that require sustained autonomous problem-solving. The models achieved 68.2% and 72.5% respectively on the SWE-bench Verified benchmark, a standard evaluation for measuring AI performance on real-world software engineering challenges. By releasing these models as open-weight, Poolside AI enables developers and researchers to access and deploy advanced coding AI capabilities. The release represents continued progress in AI-assisted software development tools.

Release of Laguna XS.2 and M.1 Models

• ▪Laguna XS.2 and M.1 are built for long-horizon tasks.
• ▪Poolside released Laguna XS.2 and M.1 as open-weight agentic coding models.

Performance on SWE-bench Verified Benchmark

• ▪Laguna XS.2 reached 68.2% on SWE-bench Verified.
• ▪Laguna M.1 reached 72.5% on SWE-bench Verified.

Perspective of Poolside AI

• ▪Poolside AI designed Laguna XS.2 and M.1 as agentic models capable of autonomous coding workflows.
• ▪Poolside AI released Laguna XS.2 and M.1 as open-weight models to enable broader developer access.

1 source