FriendliAI Expands AI Coding Model Options With Deployment of Laguna XS.2

A LinkedIn post from FriendliAI highlights that Poolside’s new open-weight agentic coding model, Laguna XS.2, is available for one-click deployment on Friendli Dedicated Endpoints. The post describes Laguna XS.2 as a 33B-parameter Mixture-of-Experts model that activates 3B parameters per token, targeting agentic coding and long-horizon development tasks.

According to the post, Laguna XS.2 is offered in four precision variants—BF16, FP8, INT4, and NVFP4—enabling customers to align model performance and memory footprint with specific GPU architectures, including Hopper and Blackwell. The company’s LinkedIn content also cites benchmark scores on SWE-bench datasets and emphasizes features such as interleaved reasoning and a 256-expert MoE design designed to keep inference efficient.

For investors, this integration suggests FriendliAI is positioning its infrastructure as a deployment layer for advanced, open-weight AI coding models, which could increase workload density and attract developer-focused enterprise users. If adoption grows, the expanded support for multiple precisions and autoscaling on dedicated GPUs may translate into higher utilization of FriendliAI’s platform and strengthen its role within the AI infrastructure and developer-tools ecosystem.