Poolside AI Introduces Laguna XS.2 and M.1: Agentic Coding Models Reaching 68.2% and 72.5% on SWE-bench Verified

Editor's take

Poolside AI has unveiled two new open-weight agentic coding models, Laguna XS.2 and M.1, demonstrating significant performance on the SWE-bench benchmark.

These models' success, particularly M.1 reaching 72.5% accuracy on SWE-bench, signals progress in developing AI agents capable of complex, multi-step coding tasks. This is crucial as the industry moves beyond single-prompt code generation towards more autonomous development tools, impacting software engineering workflows and the demand for specific developer skills. The open-weight nature also fosters broader research and development in this specialized area.

Future developments to monitor include how these models scale to even larger and more intricate software projects, and whether their performance translates to real-world development environments beyond benchmark testing. The integration of these models into existing developer platforms and the emergence of competitors offering similar long-horizon coding capabilities will also be key indicators.