118B total / 8B active per token (MoE)
1M tokens
OpenMDW-1.1
Official open-weight coding and long-horizon agent release from Poolside, positioned as the company’s most capable model to date and as a compact frontier-coding model that punches above much larger systems. The main model card and release blog describe Laguna S 2.1 as a 118B total-parameter Mixture-of-Experts model with 8B activated parameters per token, built with the same Laguna architecture family as XS 2.1: a token-choice router with softplus gating over 256 routed experts plus one shared expert, grouped-query attention, and interleaved full and sliding-window attention. Poolside says S 2.1 went from the start of training to launch in under nine weeks, supports both thinking and no-thinking modes up to 1M context, and adds broad deployment formats including BF16, FP8, NVFP4, INT4, GGUF, and a 1B DFlash draft model for speculative decoding. Release materials emphasize aggressive long-horizon coding performance at relatively low active compute, with Poolside reporting 70.2 on Terminal-Bench 2.1, 78.5 on SWE-Bench Multilingual, 59.4 on SWE-Bench Pro Public, 40.4 on DeepSWE, 46.2 on SWE Atlas Codebase QnA, and 49.7 on Toolathlon Verified, while openly publishing full evaluation trajectories for the final checkpoint. Poolside’s blog frames the release around stronger post-training, improved reasoning use, and rapid iteration in its internal Model Factory stack, while the public serving docs highlight reasoning controls, native tool-calling support, and compatibility with Transformers, vLLM, SGLang, TRT-LLM, llama.cpp, and Ollama workflows.