Poolside's Laguna S: a 118B MoE That Beats Thinky's Trillion-Parameter Open-Weights Model

AI lab Poolside has trained Laguna S, a 118B-parameter MoE model that, the company claims, outperforms the open Thinky model at roughly 1 trillion parameters. Poolside co-founder and co-CEO Eiso Kant discussed the model on the Latent Space podcast, where he described the company's internal infrastructure as a "model factory."

What is the "model factory"?

Poolside's "model factory" pairs a small team of top researchers with training infrastructure built for fast model releases. According to Eiso Kant, this pipeline let a compact team train Laguna S, and it is not the end point but the first result of a refined process.

The key idea of the approach is not to chase size for its own sake, but to build a repeatable training process in which each new model costs less and ships faster than the last.

• Developer: AI lab Poolside
• Model: Laguna S, a 118B-parameter MoE
• Claimed result: outperforms the open Thinky model (~1 trillion parameters)
• Speaker: Eiso Kant, co-founder and co-CEO
• Source: interview on the Latent Space podcast

What makes Laguna S interesting?

Laguna S stands out for the size gap: 118B parameters versus roughly 1 trillion for the open-weights Thinky model. That is almost nine times smaller, with higher quality, as Poolside claims. It gets there through a Mixture-of-Experts (MoE) architecture: each query activates only a fraction of the network's "experts," so the model runs faster and cheaper than a dense model of the same nominal size.

The gap feeds into a long-running debate over what determines model quality: raw parameter count or training engineering. Poolside is betting on the latter.

Why does this matter for the industry?

Laguna S suggests that a small team can compete with models an order of magnitude larger. According to Latent Space, the entire "factory" was built by a compact group of researchers, without the huge headcounts that market leaders rely on.

For the industry, this challenges the familiar formula of "more GPUs and more people mean a stronger model." Poolside argues that with the process set up correctly, a 118B model beats a trillion-parameter one, which means the bottleneck is shifting from scale to training engineering.

"And this is only the beginning," said

Eiso Kant, co-founder and co-CEO of Poolside, about the Laguna S release in the Latent Space interview.

What this means

Poolside says it has learned to ship competitive models with small teams, and that Laguna S is only the first in a series. If independent tests confirm the results, it will reinforce a growing trend: training efficiency and MoE architecture are starting to matter more than a model's absolute size.

Frequently asked questions

What is Laguna S?

Laguna S is a Poolside language model with a Mixture-of-Experts architecture and 118B parameters. The company claims it outperforms the open Thinky model at roughly 1 trillion parameters.

Who is Eiso Kant?

Eiso Kant is a co-founder and co-CEO of Poolside. He is the one who described the "model factory" and the training of Laguna S in the Latent Space podcast interview.