Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model That Punches Above Its Weight Class on SWE-Bench Multilingual

Poolside has released the Laguna S 2.1, a model 118B-parameter open-weight model designed for agent coding. It is a Mixture-of-Experts (MoE) model with 8B activated parameters per token. It supports a context window of up to 1M tokens in both logic and logic modes. The weights are in Hugging Face under the OpenMDW-1.1 license, and the model is small enough to run on a single NVIDIA DGX Spark.

In long-horizon coding benchmarks, the Laguna S 2.1 held its own against models several times its size, including DeepSeek-V4-Pro-Max, NVIDIA's Nemotron 3 Ultra, and Thinking Machines' Inkling. Laguna S 2.1 is a scale of the Laguna XS family, trained on the same previous training data as XS 2.1.

What is Laguna S 2.1

The model activates about 6.8% of its parameters for any given token. All 118B parameters remain in memory, but only ~8B route through the network per step. That sparsity is why a medium-sized model can behave like a large one while remaining cheap to run.

The Poolside team publishes weights for BF16, FP8, INT4, and NVFP4, as well as official modifications for GGUF and MLX and draft models for DFlash. Training began and was launched in less than nine weeks. Pre-training started on 22 May 2026 on 4,096 NVIDIA H200 GPUs. It is the first Poolside model where reinforcement learning ran with FP8 accuracy.