Poolside AI has unveiled two new agentic coding models, Laguna XS.2 and M.1, marking a significant leap in the field of automated software development. These open-weight models are specifically engineered to handle long-horizon tasks, demonstrating impressive performance on the SWE-bench Verified benchmark with scores of 68.2% and 72.5% respectively.
Designed for Complex Software Tasks
The Laguna models represent a notable advancement in the growing trend toward agentic AI systems—those capable of autonomously executing multi-step tasks. Unlike traditional models that require explicit instructions for each step, agentic models can plan, execute, and adapt their strategies in real time. This is particularly valuable in software development, where tasks often involve navigating complex codebases and solving problems that span multiple steps.
Poolside AI’s models are trained to maintain context over extended interactions, a critical feature for tasks such as debugging, feature implementation, and code refactoring. The performance scores on SWE-bench Verified—considered a gold standard for evaluating coding models—indicate that these systems are approaching human-level capabilities in software engineering tasks.
Implications for the Future of AI in Coding
These developments come at a time when the AI industry is increasingly focused on building systems that can work independently on complex, real-world problems. The success of Laguna XS.2 and M.1 suggests that agentic models may soon become the norm in developer tooling, offering solutions that go beyond simple code completion to full task orchestration.
As more organizations look to automate software workflows, Poolside AI’s models may serve as foundational tools for next-generation development environments. With continued improvements in reasoning and task execution, such models could significantly reduce the time and effort required to build and maintain software systems.
Conclusion
The release of Laguna XS.2 and M.1 underscores the rapid evolution of AI in software development. As these agentic models continue to improve, they are poised to reshape how developers approach complex coding challenges, bringing us closer to a future where AI systems can independently manage entire software projects.
