Poolside releases Laguna S 2.1 — a 118B open-weight coding model pitched as the West's answer to DeepSeek and Qwen

On 21 July 2026, Poolside released Laguna S 2.1 , an open-weight foundation model built for agentic coding that the San Francisco lab says matches or beats models several times its size. The launch drew public praise from investor Gavin Baker (@GavinSBaker), who called it "American open source FTW" and "crazy metrics for 118b parameters."

The model: A 118B total-parameter Mixture-of-Experts model with just 8B activated parameters per token, a context window up to 1M tokens, and support for thinking / no-thinking modes. It went from start of training to launch in under nine weeks and was trained in-house on 30T tokens.

Punching above its weight: Poolside calls it the most capable agentic coding model in its weight class by a wide margin. It scores 70.2% on Terminal-Bench 2.1 and 59.4% on SWE-Bench Pro (public), matching or beating models from DeepSeek, Nvidia and Tencent that carry two to eight times as many active parameters. The company concedes it is "not yet at the frontier" — closed systems from OpenAI and Anthropic still score higher.

Runs on a desktop: The model is compact enough to run on a single Nvidia DGX Spark, and the weights are on Hugging Face under the Linux Foundation's permissive OpenMDW licence. Poolside frames the release as a bet that "the West needs strong open-weight models" — noting no Western lab had released an open-weight model in the 118B class for eleven months prior.

Read more: Poolside blog · The Next Web · Gavin Baker on X