StorySoftware

Laguna S 2.1 released with 78.5% SWE-Bench Multilingual score

PoolsideLaguna Coding Models

Poolside released Laguna S 2.1, a 118-billion-parameter Mixture-of-Experts model for agentic coding that activates only 8B parameters per token. It achieved 78.5% on SWE-Bench Multilingual, topping the leaderboard while fitting on a single accelerator when quantized.

Read the original · Press

More on Laguna Coding Models

Primer