Poolside Releases Laguna-M.1, an Open MoE Model

Poolside, the startup best known for its work on AI-assisted software development, has published its first open-weights model. Laguna-M.1 arrives on Hugging Face as a mixture-of-experts (MoE) language model aimed at both general text and code generation, released under the permissive Apache-2.0 license.

The model ships with support for two of the most widely used high-throughput inference engines, vLLM and SGLang. That choice signals a focus on practical deployment: teams can run Laguna-M.1 on their own infrastructure without waiting for community patches, which lowers the barrier for evaluation and production testing alike.

Why it matters

Poolside has largely built its reputation behind a closed product, so an openly licensed release is a notable shift. A few reasons it stands out:

• Permissive licensing: Apache-2.0 allows commercial use and modification, making the model viable for internal tooling.
• MoE architecture: Mixture-of-experts designs activate only a fraction of parameters per token, a common route to balancing capability and serving cost.
• Day-one tooling: Built-in vLLM and SGLang compatibility means the model is ready to serve from the start.

Key specifics—including parameter counts, active-expert configuration, and context length—are not detailed in the release record, so prospective users will want to check the model card directly before benchmarking. As an initial release, Laguna-M.1 establishes a baseline for what could become a recurring open-model line from Poolside, and a fresh option for developers weighing open alternatives for coding workloads.