Poolside

We just crossed 10T tokens served across all our models in less than 3 months! Laguna S 2.1 is pushing new highs at ~300B tokens a day, and has processed +2T in the 14 days since release. That’s across OpenRouter, Vercel AI Gateway and our direct API. Really good to see demand for open models keep accelerating.

Poolside is joining NVIDIA Open Secure AI Alliance. We want a future where capable intelligence is available wherever defenders need it, supported by open tools that the wider community can inspect and strengthen. We help build that future by releasing our models and evaluations and contributing our research in the open. This gives defenders more access to frontier intelligence and strengthens the broader ecosystem working to make these systems secure. Learn more: https://lnkd.in/eEUyj3Bs

Today we’re introducing Poolside Desktop Assistant. It's a macOS app for running multiple coding agents across projects and repositories. The Poolside Assistant extensions bring the same experience into VS Code and Visual Studio. Coding agents have outgrown the chat box. Once agents work for hours across repositories, tools, and terminals, the developer’s role changes. It becomes less about prompting and more about directing, reviewing, intervening, and deciding what happens next. Poolside Desktop Assistant is built for that reality. It brings pool, Claude Code, Codex, and Gemini into one workspace. Run agents in parallel, isolate each task in its own Git worktree, and hand work from one agent to another without losing context. Built on ACP, the open protocol created by Zed, it keeps the interface, agent harness, and model separate and composable. Developers should not have to adopt one company’s entire stack to use a good model or agent. Out of the box, Poolside Assistant pairs Laguna S 2.1 with pool, the same harness we use internally for agent reinforcement learning and evaluation. You can also download Laguna XS 2.1 and run it fully offline through MLX. We built Poolside Assistant for ourselves and have relied on it for our daily work over the past year. Now we’re making it available to everyone. Try it on real work and tell us what gets in your way, what is missing, and what we should build next!

Today we’re releasing Laguna S 2.1, our new open-weight model for agentic coding and long-horizon work. Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model with 8B active parameters per token, up to 1M tokens of context, and thinking and no-thinking modes. It is capable enough to compete with models several times its size, yet small enough to run locally on a single NVIDIA DGX Spark. What sets Laguna S 2.1 apart is its persistence. Across long-horizon coding and research tasks, it holds onto a goal, uses tools, checks its work, recovers when an approach fails, and continues making progress for hours with little or no intervention. That persistence comes with a practical balance of cost, speed, and ownership. Laguna S 2.1 weight class is designed to take on real, long-running agentic work at a cost and speed that make it practical to run often and at scale. Because it activates only 8B parameters per token, long agent runs and reinforcement learning loops are faster and less expensive than they would be with much larger models. We’re releasing Laguna S 2.1 under OpenMDW-1.1 with checkpoints in BF16, FP8, INT4, and NVFP4, alongside official GGUF and MLX quantizations. The weights are available today on Hugging Face. Run it through pool, vLLM, SGLang, Ollama, llama.cpp, ZML, MLX, or NVIDIA TensorRT-LLM, or access it through OpenRouter and the Poolside API.

Poolside is coming to RAISE Summit. Tomorrow, Eiso Kant, Co-founder & Co-CEO of Poolside, will be speaking on two panels that sit very close to what we are building toward. AI progress is not just about better models. It is about the systems that make capability compound: training at scale, stronger infrastructure, better evaluation, and a more open ecosystem for builders to inspect, adapt, and improve the intelligence they depend on. Tune in tomorrow: Training at Scale: What a Frontier Lab Actually Needs 1:00 PM–1:40 PM – IREN AI Cloud Lounge With Shubho Sengupta, Frank Hutter, Ali Agha, and Feyza Haskaraman. From Research to Reality: Why Open Source Is the Engine of AI Innovation 4:20 PM–5:00 PM – Grace Hopper Stage With Ion Stoica, Thomas Wolf, Dillon Rolnick, Robin Rombach from Black Forest Labs, and Mark Porter. Two conversations worth being in the room for.

Today we’re releasing Laguna XS 2.1. This is a lightweight upgrade to Laguna XS.2, with stronger results on multilingual coding and terminal-style tasks. With this release, we also tried to make XS better to run locally. XS 2.1 is supported in vLLM, SGLang, NVIDIA TensorRT-LLM, HF Transformers, and Ollama. We’re releasing BF16, FP8, INT4, and NVFP4 checkpoints, so the community can choose the setup that makes sense for their hardware and deployment constraints. We’re also releasing DFlash speculator models alongside the checkpoints. In our tests, these draft models doubled achieved tokens per second, making local inference faster than before. You can try XS 2.1 now on Hugging Face, OpenRouter, via the Poolside API, Ollama, and in pool, our terminal-based coding agent.

Poolside is co-hosting Champagne, Caviar, AI & Nuggs with Modal, Gladia, turbopuffer, and Vercel during RAISE Summit ✨ Come for the technical crowd, stay for the good conversations and a menu that very much does not take itself too seriously. July 9, join us!

Frontier AI only matters if it can run where the mission happens. Today at Amazon Web Services (AWS) Summit Poolside’s Ivo Pinto will break down what it takes to bring AI-powered software development into secure environments, inside your own AWS boundary. The session will cover how teams can deploy frontier models and agents where their work already happens, from AWS to more constrained environments, and why owning the model matters for security, governance, and cost at mission scale. Session ID: AIM110-S Tuesday, June 30 | 1:30–2:00 PM Ballroom ABC – Purple Silent Stage Can’t make it? Come meet us at Booth #110

Poolside is coming to Amazon Web Services (AWS) Summit Washington, DC next week. Meet us at Booth #110 to talk about bringing AI-powered software development into secure environments, from AWS to on-premises and air-gapped deployments. Frontier AI should run where your mission requires. For government and public sector teams, this is not just about control. It is about owning durable AI capability that can be deployed, governed, audited, and operated where the mission actually happens. Poolside builds frontier models and agents that deploy inside your boundary, with the security and infrastructure requirements your mission demands. 📍 Walter E. Washington Convention Center 📅 June 30–July 1 📍 Booth #110

Today we’re releasing the weights for Laguna M.1, our most capable model to date, with a 256K context length. Both base and post-trained checkpoints are now available under Apache 2.0 licenses. M.1 is a 225B-parameter Mixture-of-Experts model with 23B active parameters, built for agentic coding and long-horizon work. Since launching M.1 through our API in April, we’ve seen strong usage on OpenRouter and through coding agents such as Kilo and Nous Research Hermes Agent. Now researchers and builders can run it on their own infrastructure, inspect it, evaluate it in their own harnesses, fine-tune it, and build on it directly. M.1 and XS.2 remain available for free on our API and through OpenRouter. We are launching dedicated paid endpoints for both models on OpenRouter for more demanding work. Open weights are now our default. We’ll keep building toward the frontier and releasing increasingly capable models in the open.