Poolside has released Laguna S 2.1, a 118B-parameter Mixture-of-Experts model designed for agentic coding tasks. The weights are available on Hugging Face under an OpenMDW-1.1 license and can run on a single NVIDIA DGX Spark.

  • It activates only 8B parameters per token while maintaining a 1M-token context window in both thinking and no-thinking modes.
  • On SWE-Bench Multilingual, it scores 78.5%, topping the published table for open disclosed-size models.
  • It achieves 70.2% on Terminal-Bench 2.1 with max thinking enabled, outperforming several larger closed systems.
  • The model was trained in under nine weeks on 4,096 NVIDIA H200 GPUs, marking Poolside's third model release in three months.

The release provides an efficient, open-weight alternative for long-horizon coding benchmarks, offering high performance at a lower inference cost compared to larger proprietary models.