Poolside Unveils Laguna Series AI Models Following Three Years and $600 Million in Funding

AIMPACT update, April 29 (UTC+8): According to monitoring by Beating, AI programming company Poolside has released the first two models in its Laguna series. Founded in 2023, Poolside has raised a total of $626 million and is valued at $3 billion, with investors including Bain Capital Ventures, NVIDIA, and eBay Ventures. Previously serving only government and public sector clients, this marks its first public model release. The flagship Laguna M.1 is a 225B-parameter MoE model (23B activated parameters), trained from scratch on 6,144 NVIDIA Hopper GPUs using 30T tokens, achieving a score of 46.9% on SWE-bench Pro and available exclusively via API. The smaller Laguna XS.2 is a 33B-parameter MoE model (3B activated parameters), scoring 44.5% on SWE-bench Pro and 68.2% on SWE-bench Verified; its weights are open-sourced under Apache 2.0 and can run locally on a Mac with 36GB of RAM, now available on HuggingFace, Ollama, and OpenRouter. Both models are available for free on a limited-time basis. Poolside has also open-sourced its internal agent runtime, pool, which uses the same environment as its training and evaluation systems. The company’s research team consists of approximately 60 members, and in its blog, it explicitly stated, “The West needs powerful open-source models—we aim to contribute.” (Source: BlockBeats)