AI & Tech

Poolside's Laguna-S 2.1 Puts a 118B Coding Model on Hugging Face

Poolside’s Laguna-S 2.1 is trending on Hugging Face: 118 billion total parameters with roughly 8 billion active per token, 256 routed experts plus one shared, 48 layers mixing global and 512-token sliding-window attention, and a one-million-token context. It is aimed squarely at agentic coding, with interleaved thinking between tool calls. Vendor-published scores are 70.2% on Terminal-Bench 2.1, 78.5% on SWE-bench Multilingual and 59.4% on SWE-Bench Pro, none independently reproduced. The licence is OpenMDW-1.1, permitting commercial use and modification — a genuinely open release in a tier where most competitors ship API access only.

Read the original — via Hugging Face ↗

← All shorts