Poolside released Laguna S 2.1, an 118B-parameter mixture-of-experts model optimized for agentic coding and long-horizon tasks. The model activates only 8B parameters per token and supports context windows up to 1M tokens, enabling complex planning and self-correction over multi-hour sessions.

Poolside reported benchmark scores of 78.5% on SWE-bench Multilingual and 59.4% on SWE-Bench Pro, outperforming models 4x to 25x larger. Weights are licensed under OpenMDW-1.1 on Hugging Face and accessible via OpenRouter and the company API, with the caveat that optimized NVFP4 quantization runs on a single NVIDIA DGX Spark. The company stated it completed development in 52 days using its internal pipeline.