Poolside: Laguna S 2.1 (free)

Poolside’s Laguna S 2.1 is designed as a coding-agent model for software-engineering tasks that involve working with code execution. The model is described as having 118 billion total parameters with 8 billion active, suggesting a large architecture that selectively activates a smaller portion for each computation. This configuration is intended to support demanding coding workflows while limiting the parameters engaged during inference.

The model is reported to score 70.2% on Terminal-Bench 2.1, a benchmark for terminal-based coding-agent behavior. That result makes it a practical fit for developers evaluating models on repository tasks, command-line operations, and iterative software work. The free variant preserves the Laguna S 2.1 model identity and is positioned for cost-conscious experimentation with coding-agent workflows.