Poolside has released Laguna S 2.1, an 118-billion parameter Mixture-of-Experts model that outperforms the 1.6-trillion parameter DeepSeek-V4-Pro-Max on the Terminal-Bench 2.1 benchmark. This American-made open model boasts a context window of up to 1 million tokens and features distinct "thinking" and "no-thinking" modes. The model was developed and released in under nine weeks and is available under the OpenMDW-1.1 license, with various quantized versions provided for local execution. AI
IMPACT Sets a new performance benchmark for open-source models, potentially influencing future development and adoption of efficient AI architectures.
RANK_REASON New model release from an AI lab with performance claims against established benchmarks. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
