Home
Poolside Launches Open Source! Laguna S 2.1 Now Free on OpenCode with Ultra-Long Context of 1M + 118B MoE Model Leads a New Era in Proxy Coding
AIbase Report: Poolside, a prominent player in AI-driven software development, has officially launched its most advanced model to date: Laguna S2.1. Now fully open source, this model is freely accessible via the OpenCode platform, generating significant buzz within the developer community.
Core Model Specifications
Scale and Architecture: Featuring an 118B parameter Mixture-of-Experts (MoE) architecture, with only 8B parameters activated per token.Context Window: Supports up to 1M tokens, making it ideal for extensive software engineering projects and complex agent workflows.Inference Modes: Offers both reasoning (thinking) and non-reasoning modes, allowing users to switch between them based on specific requirements.Performance Benchmark: Poolside asserts that Laguna S2.1 rivals much larger models in capability while remaining compact enough to operate on a single NVIDIA DGX Spark.
Open Source Availability Drives Rapid Adoption
Poolside has released the Laguna S2.1 weights under the OpenMDW-1.1 license, enabling immediate access to model files on Hugging Face. The OpenCode platform has also made the model freely available, highlighting its "1M Context · fully open source" features, which significantly lowers the barrier for developers to experiment with it.
Community feedback highlights successful deployments across diverse hardware, including Apple M3Max systems (128GB RAM) using a llama.cpp fork, as well as high-throughput inference in multi-machine DGX Spark setups. In benchmark tests, the model scored 70.2% on Terminal-Bench2.1 and demonstrated robust agent capabilities in coding challenges like DeepSWE.
Poolside's Strategic Vision
Focusing on agentic coding and long-horizon tasks, Poolside previously introduced the lightweight Laguna XS series. The launch of S2.1 signifies its shift from mid-range local models to high-performance flagship capabilities. This model combines exceptional performance with practical deployment efficiency, offering an optimal solution for local and private infrastructure.
AIbase Commentary
Related article
Tesla files in-car voice large model in Shanghai, boosting AI cockpit experience
Tesla (Shanghai) Co., Ltd. has advanced its smart cockpit localization efforts. According to a filing by Net Information Shanghai, Tesla’s vehicle voice large model service officially completed its generative AI service registration on April 20.This
Baichuan Intelligence Unveils M4 Medical AI Model for Proactive, Doctor-Like Diagnosis
Recent breakthroughs in artificial intelligence for healthcare have been achieved by Baichuan Intelligence, which has launched Baichuan-M4, a new medical-enhanced large model developed in partnership with Tsinghua University researchers. This model l
Google Gemini Unveils Auto-Execution Mode, Enabling Smartphones to Automatically Place Takeout Orders
The AI assistant is shifting from "just talking" to "doing things for you".Recently, Google has officially launched the Gemini task automation feature on Pixel 10 Pro and Galaxy S26 Ultra. This marks the official evolution of AI assistants from "conv
Related Special Topic Recommendations
Comments (0)
0/500
AIbase Report: Poolside, a prominent player in AI-driven software development, has officially launched its most advanced model to date: Laguna S2.1. Now fully open source, this model is freely accessible via the OpenCode platform, generating significant buzz within the developer community.
Core Model Specifications
Scale and Architecture: Featuring an 118B parameter Mixture-of-Experts (MoE) architecture, with only 8B parameters activated per token.Context Window: Supports up to 1M tokens, making it ideal for extensive software engineering projects and complex agent workflows.Inference Modes: Offers both reasoning (thinking) and non-reasoning modes, allowing users to switch between them based on specific requirements.Performance Benchmark: Poolside asserts that Laguna S2.1 rivals much larger models in capability while remaining compact enough to operate on a single NVIDIA DGX Spark.
Open Source Availability Drives Rapid Adoption
Poolside has released the Laguna S2.1 weights under the OpenMDW-1.1 license, enabling immediate access to model files on Hugging Face. The OpenCode platform has also made the model freely available, highlighting its "1M Context · fully open source" features, which significantly lowers the barrier for developers to experiment with it.
Community feedback highlights successful deployments across diverse hardware, including Apple M3Max systems (128GB RAM) using a llama.cpp fork, as well as high-throughput inference in multi-machine DGX Spark setups. In benchmark tests, the model scored 70.2% on Terminal-Bench2.1 and demonstrated robust agent capabilities in coding challenges like DeepSWE.
Poolside's Strategic Vision
Focusing on agentic coding and long-horizon tasks, Poolside previously introduced the lightweight Laguna XS series. The launch of S2.1 signifies its shift from mid-range local models to high-performance flagship capabilities. This model combines exceptional performance with practical deployment efficiency, offering an optimal solution for local and private infrastructure.
AIbase Commentary
Tesla files in-car voice large model in Shanghai, boosting AI cockpit experience
Tesla (Shanghai) Co., Ltd. has advanced its smart cockpit localization efforts. According to a filing by Net Information Shanghai, Tesla’s vehicle voice large model service officially completed its generative AI service registration on April 20.This
Baichuan Intelligence Unveils M4 Medical AI Model for Proactive, Doctor-Like Diagnosis
Recent breakthroughs in artificial intelligence for healthcare have been achieved by Baichuan Intelligence, which has launched Baichuan-M4, a new medical-enhanced large model developed in partnership with Tsinghua University researchers. This model l
Google Gemini Unveils Auto-Execution Mode, Enabling Smartphones to Automatically Place Takeout Orders
The AI assistant is shifting from "just talking" to "doing things for you".Recently, Google has officially launched the Gemini task automation feature on Pixel 10 Pro and Galaxy S26 Ultra. This marks the official evolution of AI assistants from "conv











