Model Releases
Laguna S 2.1: Poolside Redefines Agentic Coding with 1M Context MoE
Poolside's new Laguna S 2.1 delivers massive 1M token context and high-performance MoE architecture, setting a new standard for agentic software engineering.
July 21, 2026
Model ReleaseLaguna S 2.1
Introduction: The New Era of Agentic Coding
The landscape of AI-driven software development just shifted again. On July 21, 2026, Poolside officially released Laguna S 2.1, an agentic coding model designed to move beyond simple autocomplete and into the realm of autonomous software engineering. This isn't just another incremental update; it represents a massive leap in how models handle complex, multi-file repositories and long-running development tasks.
What makes Laguna S 2.1 stand out is its sheer velocity of innovation. Poolside has managed to ship this powerhouse in just nine weeks from the start of training to public release. This rapid iteration cycle demonstrates a highly optimized training pipeline and a clear vision for what the next generation of coding assistants must look like: highly efficient, massive context, and deeply agentic.
- Release Date: July 21, 2026
- Focus: Agentic software engineering and repository-level reasoning
- Development Speed: Shipped within 9 weeks of training start
Architecture: The Power of Mixture-of-Experts (MoE)
At the heart of Laguna S 2.1 lies a sophisticated Mixture-of-Experts (MoE) architecture. The model boasts a massive 118B total parameter count, but thanks to its MoE design, only 8B parameters are active during any single inference pass. This allows the model to retain the deep knowledge base of a massive LLM while maintaining the lightning-fast latency and lower compute requirements of a much smaller model.
This architectural efficiency is a significant upgrade over the previous Laguna M.1. Laguna S 2.1 provides substantially stronger reasoning capabilities while running at roughly half the active size of its predecessor. This balance of 'intelligence per watt' is critical for developers integrating these models into real-time IDE workflows and complex CI/CD pipelines.
- Total Parameters: 118B
- Active Parameters: 8B (MoE)
- Architecture: Mixture-of-Experts for optimized inference
- Efficiency: Higher performance than M.1 at half the running size
Performance & The Scaling Roadmap
The development of Laguna S 2.1 follows a rigorous validation process. Before scaling up to this 118B version, Poolside released the Laguna XS 2.1, which successfully matched the M.1 model on the SWE-bench Multilingual benchmark despite being only a seventh of the size. This validated the 'recipe'—the specific training data mix and optimization techniques—that was then scaled up to create the S 2.1 powerhouse.