Laguna XS 2.1

Laguna XS 2.1 is a lightweight AI coding model developed by Poolside, designed for Agentic Coding, local AI development environments, and efficient software engineering workflows.
Compared with Laguna S 2.1, XS 2.1 focuses more on efficiency, flexible deployment, and running AI coding capabilities on personal or resource-constrained environments.
Built with a Mixture-of-Experts (MoE) architecture, the model features approximately 33B total parameters with around 3B active parameters per token, reducing computational requirements while maintaining strong coding performance.
Laguna XS 2.1 supports up to 256K-token context windows, allowing developers to analyze large codebases, understand complex project structures, handle terminal-based tasks, and manage long-running software development workflows.
With support for local deployment through platforms such as vLLM, SGLang, TensorRT-LLM, and Ollama, Laguna XS 2.1 provides developers with a flexible open-weight solution for AI coding assistants, automated development tools, and private AI programming environments.