US startup Poolside debuts its first open-weight model, Laguna XS.2, a 33B-A3B-parameter MoE model, and Laguna M.1, a proprietary 225B-A23B-parameter MoE model
The AI race lately has felt a bit like a game of tennis: first, Anthropic releases a new, pricey state-of-the-art proprietary model …
VentureBeatCarl Franzen
Context & Ripple Effects
Poolside’s release arrives amid a broader expansion of open-weight alternatives from startups and research groups, including Arcee’s Trinity Large and Ai2’s Olmo 3, while smaller open-weight releases such as Alibaba’s Qwen3.5 series broaden the range of deployment options.
Related coverage also shows U.S. startups adopting Chinese open-weight models for cost and customization. Poolside is positioning its own open-weight offering alongside a much larger proprietary model, rather than choosing one distribution model exclusively.
First-order effects
Poolside now offers customers an open-weight Laguna XS.2 model they can evaluate and customize, alongside proprietary access to Laguna M.1.
The release gives Poolside two distinct routes to adoption: open-weight distribution for XS.2 and a controlled product path for M.1.
Second-order effects
Model buyers gain another non-closed option, increasing pressure on both open-weight developers and proprietary providers to justify differences in capability, customization, and commercial terms.
A dual open/proprietary lineup may make procurement more comparative: organizations can test an open model for fit while reserving proprietary models for workloads where they see a clear advantage.
Third-order effects
If more startups pair open-weight releases with proprietary frontier models, AI model competition may settle into a segmented market: broadly deployable models establish reach, while closed models seek to monetize differentiated performance or service.
The growing supply of credible open-weight models could shift durable leverage toward buyers and deployers, though that depends on whether they can operate, adapt, and support those models effectively.
The trend: AI vendors are increasingly treating open weights and proprietary models as complementary distribution and monetization layers rather than mutually exclusive strategies.
Today we're releasing Laguna XS.2, Poolside's first open-weight model. It's a 33B total / 3B active MoE model built for agentic coding and long-horizon tasks. Trained fully in-house on our own stack. Runs on a single GPU. Released under Apache 2.0. Links 👇 Weights: [image]
Today we're launching our first public Poolside models: Laguna M.1 and Laguna XS.2, and we've built ❈Shimmer, an instant-on VM sandbox with Poolside Agent pre-installed so you can try them out. Go play out our new models for free, and build something fun → [video]
We believe software is the interface through which intelligence becomes useful, and the pathway to true intelligence Until today we've focused on building the full stack by shipping models, the Poolside platform, and our applications to very large enterprise customers, on-prem
There should be more noise about Poolside models They pre trained them from scratch and not fine tuned on other OSS models Beats Gemma models and on par with Qwen Western (US) model that has a small on device model and large through Cloud https://poolside.ai/models [image]
Day-zero support for Laguna XS.2 in MLX🔥🚀 @poolsideai's first open-weight model is now supported in MLX. 33B total params, 3B activated, built for agentic coding, and running natively on Apple Silicon. Huge thanks to team at Poolside for the early collaboration 🙌🏽 Heads up:
We'll be making public all our applications, including our VSCode IDE assistant. For now you'll simply have to make do with what I think is likely the best web based instant-on VM sandbox experience available Haven't been this in love with a product since “git push heroku
Laguna M.1 is a 225B MoE model with 23B active. Poolside's most capable model, purpose-built for agentic coding and long-horizon work. Use it now: https://openrouter.ai/...
I'm particularly excited for Laguna XS.2, our open weight model you can run on your own hardware. This thing is smart and fast. Use it for free in our sandbox env https://shimmer.run/ with a Next.js app skeleton for maximum fun. [video]
@zzddfge @poolsideai Different training targets Laguna is focused on coding and long-horizon tasks. While Qwen 3.6 is great an all-rounder and a VLM (the multimodality helps with generalisation )
🚀 Poolside (🇺🇸) just dropped Laguna XS.2, their first open-weight model! Test it now on @OpenRouter for Free! 🔥 33B total / 3B active MoE model 🤖 Built for agentic coding & long-horizon tasks 💻 Runs on a single GPU ✅ Apache 2.0 license Key highlights: • 📈 Strong [image]
Today @poolsideai is releasing Laguna M.1 & Laguna XS.2, our latest generation models and first public models We started Poolside because we believed that to build truly capable coding agents, you need to own the full stack: data, training, reinforcement learning, inference.
Laguna XS.2 is open-weight under Apache 2.0 and runs on a single GPU We believe the West needs strong open-weight models. We want developers and researchers to be able to run this one, build on it, and push it further than we can on our own Laguna M.1 is our most capable model
Big day! Incredible what Eiso Kant, Jason Warner, Margarida Garcia, José Caldeira and the team at Poolside have accomplished in a short amount of time. …