Poolside Releases 118B Laguna S 2.1, Matching Larger Coding Models on One Desktop
Updated
Updated · The Manila Times · Jul 22
Poolside Releases 118B Laguna S 2.1, Matching Larger Coding Models on One Desktop
3 articles · Updated · The Manila Times · Jul 22
Summary
Laguna S 2.1 is available now under an OpenMDW-1.1 license, with Poolside positioning the 118B open-weight model as a self-hostable option for agentic coding.
On Terminal-Bench 2.1 and SWE-Bench Pro, the model scored 70.2% and 59.4%, matching or beating larger rivals including DeepSeek-V4-Flash, Nemotron 3 Ultra and Inkling.
A single NVIDIA DGX Spark can run the model locally, letting enterprises and regulated users shift coding agents off metered APIs while keeping code and data in-house.
Poolside said Laguna S 2.1 was trained end-to-end in under 4 weeks on 4,000 H200 GPUs, using its automated Model Factory to support roughly five-week release cycles.
The launch targets a gap in Western open-weight AI, where many leading downloadable models now come from Chinese labs and no Western lab has released this size class in 11 months.
A new AI model rivals tech giants but runs on one PC. Is the cloud AI monopoly over?
After its $14B AI campus collapsed, can Poolside's new desktop model revive its fortunes?
Poolside’s Laguna S 2.1 (118B): Open-Weight AI Model Sets New Benchmark for Coding and Transparency
Overview
Poolside has launched Laguna S 2.1, a powerful open-weight AI model with 118 billion parameters and a Mixture-of-Experts architecture, aiming to set new standards in coding and AI performance. Driven by co-founder Eiso Kant’s vision that intelligence should become a commodity, Poolside is committed to making open models as strong as, or stronger than, closed-source alternatives. This strategic release highlights Poolside’s ambition to push the boundaries of open AI, offering advanced capabilities and efficiency while fostering an open ecosystem for innovation and accessibility.