Poolside says its new open-source coding model beats a 1.6T-parameter DeepSeek

Poolside, the company building AI agents for programming, has open-sourced its latest coding model: Laguna S 2.1, a mixture-of-experts (MoE) design with 118 billion total parameters. The trick? It only activates 8 billion parameters per token, and it supports a 1 million-token context window. After quantization, the whole thing can run on a single NVIDIA DGX Spark.

In Poolside's own benchmarks, Laguna S 2.1 scored 70.2% on Terminal-Bench 2.1 — higher than DeepSeek-V4, which has 1.6 trillion total parameters and 49 billion active parameters…