Poolside Releases 118B Open-Weight Coding Model Laguna S 2.1

AI coding company Poolside has released Laguna S 2.1, its latest open-weight coding model. Designed specifically for agentic coding and long-horizon tasks, the model quickly topped the Hugging Face trending charts thanks to its ultra-long context support and strong benchmark performance.

Confirmed

Laguna S 2.1 utilizes a Mixture-of-Experts (MoE) architecture with 118B total parameters and 8B active parameters per token. It supports a context window of up to 1 million tokens and offers both thinking and no-thinking modes. Released under the OpenMDW license, it is built on the Transformers architecture and supports vLLM deployment. Poolside emphasized the model's stable performance in long-chain tasks and its ability to outperform some much larger models in agentic coding. Officially, it scored 70.2 on the Terminal-Bench 2.1 benchmark, rivaling or surpassing models 5 to 25 times its size. Thanks to its low active parameters, the model runs efficiently on a single NVIDIA DGX device and has day-0 support on SGLang. According to @benburtenshaw, it can already run locally in llama.cpp via a fork, allowing users to directly load quantized files. Regarding the development timeline, @Madisonkanna noted it took only 8 weeks, while @nathanbenaich mentioned a 9-week period.

Why it matters

The release of Laguna S 2.1 is more than just a new model launch; it sparks a deeper discussion about the "US open-source AI approach" and the economics of open weights. In a related interview, @Madisonkanna pointed out that this touches upon the larger question of "who is qualified to build intelligence." Its exceptional local inference efficiency also provides significant convenience for developers in actual deployment and testing.

2026-07-22 ~ 2026-07-23 · 35 related posts

Full story(2 episodes)→

• Episode 1: Poolside Releases 118B Open-Weight Coding Model Laguna S 2.1(2026-07-22, 35 posts)
• Episode 2: Poolside's Laguna S 2.1 Available for Free on Nous Portal(2026-07-22, 3 posts)

• Poolside launches Laguna S 2.1 with 118B parameters and 8B active per token — Madisonkanna · 2026-07-22
• Poolside’s Laguna S 2.1 launches as a 118B open-weight MoE with 1M context — NVIDIAAI · 2026-07-22
• Poolside’s laguna s 2.1 claims 118B MoE, 1M context, and 78.5% on SWE-bench Multilingual — ben_burtenshaw · 2026-07-22
• Poolside releases Laguna S 2.1, a 118B open-weight coding agent with 1M context — ben_burtenshaw · 2026-07-22
• Poolside launches Laguna S 2.1, a 118B MoE with a 1M-token context window — Lowkey_LokiSN · 2026-07-22
• [source] Poolside releases Laguna S 2.1, a 118B open-weight coding model with 1M context — Every-Walrus · 2026-07-22
• Poolside says Laguna S 2.1 is a 118B open-weight agentic coding model with 1M context — GavinSBaker · 2026-07-22
• Poolside Releases Laguna s 2.1: A 118B Local Coding Agent Model — ben_burtenshaw · 2026-07-22
• Poolside Releases Laguna S 2.1: Small Parameters Rivaling Larger Models — cephaloform · 2026-07-22
• [source] Poolside Releases Laguna S: An 118B Parameter Open-Source Model — Madisonkanna · 2026-07-22
• Poolside’s Laguna S launch highlights a 118B open-weight model with 1M context — Madisonkanna · 2026-07-22
• Poolside’s Laguna S interview focuses on open-weight economics and model design — Madisonkanna · 2026-07-22
• Poolside Open-Sources Laguna S Model, Discusses Software Engineering and Open Weight Economics — baseten · 2026-07-22
• [source] Poolside Releases Laguna-S-2.1 Model, Tops Hugging Face Trending — poolside · 2026-07-22
• Poolside opens Laguna S 2.1, a 118B coding model with 1M context — AccBalanced · 2026-07-22
• Poolside Releases Laguna S 2.1: A Small Coding Model Outperforming 1T+ Parameter Giants — mervenoyann · 2026-07-22
• Poolside details Laguna S 2.1 base as a 118B MoE with 8B active params — cephaloform · 2026-07-22
• Poolside’s Laguna S launches with 118B parameters, 8B active and 1M context — Madisonkanna · 2026-07-22
• Poolside Releases 118B Laguna S 2.1, Outperforming 1T+ Flagships in Coding — airesearch12 · 2026-07-22
• Poolside releases Laguna S 2.1, a 118B open model with 1M-token context — willccbb · 2026-07-22
• Poolside Releases Laguna S 2.1: 118B MoE Model Runs on a Single DGX — joecole · 2026-07-22
• Poolside releases Laguna S 2.1 with 118B parameters and long-horizon coding focus — realsohamparekh · 2026-07-22
• Poolside releases Laguna S 2.1, a 118B coding-agent model with 1M context — burny_tech · 2026-07-22
• Poolside releases 118B open Laguna S 2.1 for single-node local deployment — IanAndrewsDC · 2026-07-22
• Laguna S 2.1 packs 118B parameters, 8B active, and a 1M-token context — UsedMorning9886 · 2026-07-22
• Poolside’s Laguna S 2.1 brings 118B parameters and 1M context to a single DGX Spark — ying11231 · 2026-07-22
• Poolside launches Laguna S 2.1, a 118B open-weight MoE with 1M context — Dan_Jeffries1 · 2026-07-22
• Poolside says its 118B-A8B open-weight model hits 70%+ on TB2.1 in under nine weeks — nathanbenaich · 2026-07-22
• Poolside’s open-weight Laguna S 2.1 reportedly beats coding rivals 10x larger — MichaelFNunez · 2026-07-23
• Poolside’s Laguna S 2.1 brings 118B open weights, 1M context and 8B active parameters — WorldofAI · 2026-07-23

5 near-duplicate retellings: Madisonkanna · baseten · Madisonkanna · ivan_bezdomny · nathanbenaich