Laguna-S-2.1: Open-Source MoE Model by Poolside for Agentic Coding

LinkPoolsideTool

Poolside has released Laguna-S-2.1, a Mixture-of-Experts language model featuring 118B total parameters with approximately 8B activated parameters per token. Targeted at software engineering and long-horizon agentic workflows, the model is available under the permissive OpenMDW-1.1 license.

Key points

  1. Architecture: MoE model with 118B total parameters (~8B active per token), 256 routed experts (top-10) plus one shared expert, across 48 layers.
  2. 1M Context Window: Handles up to 1,048,576 tokens, employing 12 global and 36 sliding-window attention layers.
  3. Native Reasoning: Supports interleaved thinking before and between tool calls, configurable on a per-request basis.
  4. Inference Optimisations: Supports speculative decoding via an accompanying DFlash draft model, alongside FP8, NVFP4, INT4, and GGUF quantizations.
  5. Ecosystem Compatibility: Ready for deployment with vLLM, SGLang, TRT-LLM, llama.cpp, Ollama, and Hugging Face Transformers.
  6. OpenMDW-1.1 License: Permissive terms allowing free commercial and non-commercial deployment and modification.

Why it matters

Laguna-S-2.1 provides development teams with an open-weights MoE model capable of processing a 1M token context while keeping compute per token relatively low due to activating only 8B parameters. Its native tool-use and reasoning capabilities provide a flexible alternative to closed proprietary models for agentic coding.

Analysis

Poolside places Laguna-S-2.1 between its smaller XS and larger M tiers. Evaluated results—including 78.5% on SWE-bench Multilingual and 70.2% on Terminal-Bench 2.1—reflect its optimisation for code reasoning and agentic pipelines. While serving the full BF16 checkpoint (~236GB) requires multiple GPUs, quantization options and native integration into frameworks like vLLM, SGLang, and Ollama make it broadly accessible.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Link:Kwaipilot

    KAT-Coder-V2.5-Dev: Open-Weight MoE Model for Coding Agents

    Kwaipilot has released KAT-Coder-V2.5-Dev, an open-weight Mixture of Experts model based on Qwen3.6-35B-A3B. Featuring 35 billion total and 3 billion activated parameters, it has been specifically post-trained for agentic software engineering tasks.

    AI & LLMs· Tool

  • Link:ornith-ai

    Ornith-1.5-397B: Open-Source MoE Model Focused on Coding and Reasoning

    ornith-ai has released Ornith-1.5-397B, an open-source multimodal mixture-of-experts model under the MIT licence. With 397 billion parameters and an end-to-end self-improvement loop, it delivers coding and reasoning benchmark performance comparable to top proprietary models.

    AI & LLMs· Resource

  • Link:Z.ai

    Z.ai Releases GLM-5.2 with 1M Context Window and MIT Licence

    Z.ai has released GLM-5.2 on Hugging Face under the permissive MIT licence. The flagship open-weights model is designed for long-horizon agentic and coding workflows, featuring a robust 1M-token context window. Architectural updates including IndexShare and an enhanced MTP layer significantly cut computation FLOPs and accelerate speculative decoding.

    AI & LLMs· Resource

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.