Poolside Laguna: A small model that outperforms trillion parameter models
The Poolside team just dropped Laguna S 2.1 — a 118 billion parameter model that only uses 8 billion parameters at once. It beat trillion param models on the DeepSWE benchmark for AI developers and solved Erdős Problem #397 from 1975 in 40 steps for just $0.09!
The Poolside team recently released a new code generation model called Laguna S 2.1. This open source code generation model has 118 billion parameters, but only 8 billion are actually being used during inference. Not only did this model outperform trillion parameter models on the DeepSWE benchmark for AI developers (scoring 40.4%), but it also solved Erdős Problem #397 (unsolved since 1975) in just 40 steps for only $0.09!
They were able to do this by training the model to be “persistent,” meaning that it would check its own work, explore alternative approaches when things weren’t going well, etc. In other words, training methods matter too! You don’t always have to race to reach a trillion parameters.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles