225.8B total / 23.4B active (MoE)
256K tokens
Apache 2.0
Official open-weight flagship coding model in Poolside's Laguna family, positioned for long-horizon agentic software engineering and terminal work. Poolside describes Laguna M.1 as a sparse Mixture-of-Experts transformer trained completely from scratch in-house on more than 30T tokens using its internal 'Model Factory' stack, with 225.8B total parameters and 23.4B activated per token. The technical report and release blog emphasize that Laguna M.1 finished pretraining first and served as the foundation from which Poolside derived later Laguna-family lessons, while the open-weight release now includes post-trained and base checkpoints plus efficiency-oriented variants in the public Hugging Face collection. Official materials highlight large-scale pretraining on 6,144 interconnected NVIDIA Hopper GPUs, heavy synthetic-data use across the Laguna family, a distributed Muon optimizer implementation, and a fully asynchronous online agent-RL system built around Poolside's own coding-agent harness. Poolside reports benchmark results of 74.6 on SWE-bench Verified, 63.1 on SWE-bench Multilingual, 49.2 on SWE-Bench Pro, and 45.8 on Terminal-Bench 2.0 in the technical report, with all agentic evaluations run at 256K context using its internal harness and patched benchmark images to address infrastructure and reward-hacking issues. The Hugging Face collection and model cards also indicate broad open deployment support through base and post-trained checkpoints and multiple public formats, including BF16, FP8, and NVFP4 variants, under Apache 2.0 licensing.