Poolside Releases Laguna S 2.1 Agentic Coding Model

Poolside released Laguna S 2.1, an open-weight, 118B-parameter agentic coding model, on May 22, 2026. This Mixture-of-Experts (MoE) model activates approximately 8 billion parameters per token, allowing it to perform comparably to much larger models while maintaining efficiency. Laguna S 2.1 supports a context window of up to 1 million tokens and is designed for long-horizon coding tasks.

The model's weights are available on Hugging Face under an OpenMDW-1.1 license, and it is optimized to run on a single NVIDIA DGX Spark. Poolside has published weights in various precisions, including BF16, FP8, INT4, and NVFP4, along with GGUF and MLX conversions. The development cycle from the start of training to launch was completed in under nine weeks.

Pre-training for Laguna S 2.1 commenced on May 22, 2026, utilizing 4,096 NVIDIA H200 GPUs. This release marks the first Poolside model where reinforcement learning was conducted using FP8 precision. The model is a scaled-up version of the Laguna XS family, trained on the same pre-training data as Laguna XS 2.1.

In performance evaluations, Laguna S 2.1 achieved a score of 70.2% on Terminal-Bench 2.1 with thinking enabled, positioning it as the leading open-weight model of disclosed size on Poolside's leaderboard. On the SWE-Bench Multilingual benchmark, the model scored 78.5%, outperforming all other published models.