<cite index="21-2">Poolside released Laguna S 2.1, a 118-billion-parameter Mixture-of-Experts (MoE) system that activates only 8 billion parameters per token, supports a context window of up to 1 million tokens</cite>, and <cite index="21-2">matches or beats open models several times its size on agentic coding tasks</cite>. <cite index="21-2">The weights are available immediately on Hugging Face under the permissive OpenMDW-1.1 license</cite>.
<cite index="21-4">Laguna S 2.1 scores 70.2% on Terminal-Bench 2.1, placing it 11th on the company's compiled leaderboard — ahead of DeepSeek-V4-Pro-Max, a 1.6-trillion-parameter model that scored 64.0</cite>. <cite index="21-4">On SWE-Bench Multilingual, it posts 78.5%, and on SWE-Bench Pro's public dataset, 59.4%</cite>. <cite index="22-5">Laguna S 2.1 took less than 4 weeks to train end to end on 4,000 H200 GPUs</cite>.
<cite index="21-5">Poolside's core business is deploying models inside the security boundaries of government, defense, and regulated enterprises — customers for whom closed, metered API access is often not acceptable</cite>. For teams running agentic coding workloads, the ability to self-host a 118B MoE model on a single DGX Spark removes API-dependency friction; the open-weight license and reproducible benchmarks matter more in this market segment than frontier leaderboard position.