Poolside released its most capable model, Laguna S 2.1, on Tuesday, a 118-billion-parameter Mixture-of-Experts system that activates only 8 billion parameters per token and supports a context window of up to 1 million tokens. The open-weight model, available immediately on Hugging Face under the OpenMDW-1.1 license, matches or beats open models several times its size on agentic coding tasks. Poolside reports it scores 70.2% on Terminal-Bench 2.1, placing 11th on the company's leaderboard ahead of DeepSeek-V4-Pro-Max (a 1.6-trillion-parameter model scoring 64.0), Thinking Machines' 975-billion-parameter Inkling (63.8), and Nvidia’s 550-billion-parameter Nemotron 3 Ultra (56.4). On SWE-Bench Multilingual, it posts 78.5%, and on SWE-Bench Pro's public dataset, 59.4%. The model went from pre-training start on May 22 to public launch in under nine weeks, trained on 4,096 Nvidia H200 GPUs. Poolside frames the release as a response to the West's open-weight AI gap, noting no Western lab has released open weights in this size class in 11 months since OpenAI's gpt-oss-120b last August.