Poolside just dropped Laguna S 2.1, a massive 118B parameter agentic model. Utilizing a hyper-efficient Mixture-of-Experts architecture, it activates only 8B parameters per token.

With a 1M context window and reinforcement learning, this beast dominates multilingual coding tasks. It rivals industry giants while running smoothly on a single NVIDIA DGX Spark.

The future of autonomous software engineering has officially arrived.