Poolside released Laguna S 2.1, a 118B-parameter mixture-of-experts model with 8B activated parameters and a 1M context window. The model runs on a single DGX Spark and is available on Hugging Face.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
NVIDIA AI officially congratulates Poolside on the release of Laguna S 2.1, an open-weight MoE model with 118B total parameters, only 8B activated per token, supporting a 1M context window, and featuring thinking and non-thinking modes.
On July 21, 2026, Poolside announced the release of Laguna S 2.1, calling it "its most powerful model to date." The model adopts a Mixture of Experts (MoE) architecture with 118B total parameters, but only activates 8B parameters per token, aiming to balance performance and efficiency.
The model supports a context window of up to 1M tokens and features both thinking and non-thinking modes to adapt to different task requirements.
Laguna S 2.1 is an open-weight model, available for download on Hugging Face, and can be tried directly on OpenRouter. NVIDIA AI congratulated the Poolside team on X, noting that the model "performs far beyond its size" and "runs well locally."
Additionally, the model supports customization via NVIDIA NeMo, further expanding its application scenarios.
This report is based on Poolside's official X account announcement and NVIDIA AI's retweet confirmation. All specifications (118B total parameters, 8B activated, 1M context, thinking mode) come from the original statement and are at the source_claim level.
Laguna S 2.1 is an open-weight MoE model featuring 118B total parameters, 8B activated, and 1M context, balancing performance and efficiency, and supporting local operation and customization.
Primary report
Primary source
Also covered by 2 sources