Poolside, the AI coding startup valued at $3 billion after raising $500 million in 2024, has released a new open-weight model that it claims can outperform much larger rivals on agentic software development tasks.
Laguna S 2.1 has 118 billion parameters, but activates only eight billion for each token it processes. Poolside says the model is capable of handling long-running coding tasks while remaining efficient enough to operate on a single Nvidia DGX Spark system.
“It does the work of models several times its size on agentic coding,” Poolside Co-CEO and Co-Founder Eiso Kant wrote on X (a serial European founder), describing the model as “far more capable than anything we have created before”.
Public benchmarks reportedly place Laguna ahead of most comparable American and Chinese open-weight models, although it still trails Moonshot AI’s Kimi K3 (the Chinese model which has been making waves) in some tests.
Poolside co-founder and co-CEO Jason Warner told Forbes that Laguna was the result of 18 months spent developing an industrialised “model-building factory”, which the company believes will allow it to release increasingly capable models every five weeks.
“The classic notion of model building is an artisanal process,” Warner said. “We’ve built an industrial model-building process.”
For his part, Kant positioned the release as part of a broader attempt to prevent advanced AI from becoming concentrated among a handful of closed-model companies.
“The open ecosystem will not win by being the best in its own category,” he wrote. “Open models have to be on par with, or better than, their closed equivalents.”
Laguna S 2.1 is available under the OpenMDW-1.1 licence, with its weights published on Hugging Face and access offered through OpenRouter and Poolside’s API.
