Woofun AI reports that Poolside has open-sourced Laguna S 2.1, a Mixture of Experts model containing 118 billion total parameters with 80 billion activated per instance. The model supports a one-million-token context window and can operate on a single NVIDIA DGX Spark after quantization.

In internal evaluations, Laguna S 2.1 scored 70.2% on Terminal-Bench 2.1, exceeding DeepSeek-V4-Pro-Max's 64.0% score. It also surpassed benchmarks on SWE-Bench Multilingual, SWE-Bench Pro, and DeepSWE.

However, it trails Kimi K3, which recorded 88.3% on Terminal-Bench 2.1 and 69% on DeepSWE. Poolside notes these results were generated using its proprietary Agent framework, unlike competitors' publicly reported scores.