🚀 The Laguna Model Family
Laguna M.1 (Flagship): A massive 225B parameter Mixture-of-Experts (MoE) model with 23B active parameters. Trained from scratch on 30 trillion tokens using 6,144 NVIDIA Hopper GPUs, it achieves a 46.9% score on SWE-bench Pro. This model is available exclusively via API.
Laguna XS.2 (Efficient): A lightweight 33B parameter MoE model (3B active). It delivers impressive performance with a 44.5% score on SWE-bench Pro and 68.2% on SWE-bench Verified. Crucially, Poolside has open-sourced the weights under the Apache 2.0 license. Optimized for efficiency, it can run locally on Macs with 36GB of RAM and is already available on HuggingFace, Ollama, and OpenRouter.
Comments & Questions (0)
No comments yet
Be the first to comment!