Today we’re releasing Poolside Laguna S 2.1 It is a 118B-total, 8B-active open-weight model built for agentic coding and long-horizon work, with context up to 1M tokens poolside.ai/blog/introducing… Laguna S 2.1 sits at the top of its weight class and competes with open models many times larger, while remaining small enough to run on a single NVIDIA DGX Spark It is available today under OpenMDW-1.1 via Hugging Face, OpenRouter, the Poolside API, and pool This is a remarkable model by any measure. As much as we can gather, it's the best open weight model in the West, regardless of size But the model itself is only part of the story There is a prevailing narrative that building capable models requires ever more capital, compute, and people. We believe the more important question is how efficiently you can turn those resources into intelligence At Poolside, we approach model building as an industrialized process spanning data, pre-training, reinforcement learning, evaluation, and inference. We call that system the Model Factory. We have written about our approach to model building extensively in a 6 part blog series: poolside.ai/blog/introducing… Laguna S 2.1 went from the Model Factory kicking it off to release in 52 days. The model is the output. The ability to keep building better models, faster and more efficiently each time is the actual innovation. The Model Factory is Poolside's compounding asset And so here we are, 52 days after kicking this off, releasing a 118B/8B MOE that tops 70% on TB 2.1, 78% on SWE-Bench Multi, 59% on SWE-Bench Pro, and 40% on DeepSWE. And it's fully open, from an American company I grew up in the Linux and Python communities, then spent much of my career at Canonical (Ubuntu), Heroku and GitHub. Those communities shaped my belief that important technology becomes more useful when people can understand it, challenge it, and build upon it. And even more important than being useful is being trusted. That is what open weights mean to me. They are not a marketing or distribution exercise. They give people control: the ability to inspect the work, reproduce the claims, modify the model, and run it inside their own environment I believe the West needs a credible open path to frontier intelligence. I believe it should be from an American company. We intend to be that company Laguna S 2.1 is another step in that direction and yet more evidence of Poolside's long held, often times contrarian beliefs and views on how intelligence will be built. One last note. We are doing something very different that we hope becomes industry norm going forward. We recognize that releasing a 118B/8B MOE that performs as well as Laguna S does would be met with some degree of skepticism. Models of this size are not supposed to outperform models 4-25x larger. We double and triple checked our benchmark trajectories to be sure. But we wanted to go another step and release those trajectories for you to see and help us quadruple check them. trajectories.poolside.ai/ If you find something we missed, we genuinely want to hear about Have fun building whatever thing you can think of with the most persistent little model that could

*cough cough* The Model Factory(tm) *cough cough* Once people fully understand the real math of models, things will finally start to click. Not a lot of folks understand why we did what we did and have gone the route we did. Dylan clearly does (as he does on the entire AI stack). And since the launch of Laguna series of models, the "wait, how these jokers just do that?" sort of questions have started to shine a light a bit more. Still much more to come. The full Model Factory blog series: poolside.ai/blog/introducing…

March 2024 (11 months after starting poolside) I woke up at 4am and wrote down 2 hours of thoughts on models and where LLMs should be going. Interesting to see how much of it has played out since. Shared back then only with @PengmingWang, @NikolayZinov and @pmarca 1/🧵

Fun little update. Still climbing the charts.

Get in touch with @badiaserra to bring Laguna S to your apps And some fun tidbits from the past 2 weeks since S 2.1 launch (from a internal dashboard Andrea and our amazing design team built to track it all) Creeping up on 100k downloads/day on HF Top 10 open model (gaining 7 spots this week), and top 11 all models (gaining 23). As far as western open models, it's Nvidia and Poolside and that's it

Native steering + queueing. You can now give an agent a quick “keep going, but by the way…” while it’s working, without interrupting the task. Or queue up several prompts for it to work through next. Available wherever the harness supports it.

Poolside Desktop Assistant has been out for a week, and we’ve received so much great feedback! Today we’re releasing version 1.4.0, with a bunch of new features and fixes based on your bug reports: - Native steering and queueing where supported - Proper plan mode and agent Q&A - First-class subagents, with full transcripts for Claude and better status reporting for Codex - Much faster inference for local models - A whole load of smaller bug fixes More below!

Demand for Laguna S 2.1 has been ridiculous (in the good way). Team has been working to meet that demand through a variety of methods. Please enjoy more Laguna now and have a fun weekend hacking away on whatever your building.

We've improved our serving efficiency and increased rate limits by +10x. The update is live on @OpenRouter @vercel AI Gateway and platform.poolside.ai. Thank you to everyone who has used the model, shared feedback and stuck with us while we improved the experience! Usage is already climbing. On OpenRouter alone, Laguna S 2.1 is on pace for ~250B tokens today, 4x our daily average this week. We are also taking 10% off our paid endpoint on OpenRouter. It's a dedicated deployment with the full 1M context window for the best performance on harder tasks. Run Laguna S 2.1 in pool or Poolside Desktop Assistant, or plug it into @opencode, @NousResearch Hermes Agent, @kilocode, @cline or @pidotdev and let it run over the weekend. We'll be watching the graphs.

Digging the @poolsideai desktop app. The craft is evident. Laguna model is also doing a great job of tackling a fairly sized non-trivial monorepo. Hats off to @jasoncwarner @eisokant and team.

The entire design team at Poolside is amazing In particular I am happy that the world can finally see all the foundations and groundwork that @almonk and @dizzyup did as our first founding design engineers, and that's before mentioning them recruiting the rest of the amazing group of humans Impeccable taste and talent all around

Love to see it

What I really like about this release is that it is fully built around ACP (Agent Context Protocol). That means Bring Your Own Agent & Bring Your Own Model. You can use multiple agents, handover from one to the other, no lock in whatsoever. This has quickly become the daily driver for quite a few folks on our team. Give it a spin!!

Poolside Desktop Assistant is available today for macOS, VS Code, and Visual Studio. Download it and put it to work: poolside.ai/downloads/deskto… Learn more about why we built it: poolside.ai/blog/introducing… Try it now and tell us what to build next in Discord: discord.gg/NAnCKRbZa

A few of the things we love about this product: You can bring pool, Claude Code, Codex, or Gemini into one workspace and customize the layout around how you work. Run agents in parallel, each isolated in its own Git worktree, so they never step on each other. Hit a limit? Hand the session off to another agent and keep going, context stays the same.

Today, we’re releasing Poolside Desktop Assistant. One place to run coding agents across macOS, VS Code, and Visual Studio. We built it for ourselves and have used it every day for the past year. Now we’re opening it up to everyone.

Today we’re releasing the Poolside Desktop Assistant, one workspace for running coding agents across macOS, VS Code, and Visual Studio We've all got this incredible new intelligence layer and yet most of our interfaces into this remarkable thing are still "I can fit so many chat interfaces onto this bad boy" thinking. And just like we did away saying "Gen AI", we can move past chat box exclusive thinking Agents are already pushing tens of hours working for us boundaries, and with new long-horizon optimized models like Laguna S 2.1, all of us can and should move beyond question/answer style back and forth with agents...and so our tools should evolve too Poolside Desktop Assistant is model and harness agnostic, with Laguna S 2.1 + pool as the default. It supports any ACP-compatible agent, so go nuts (though we think you'll fall in love with S 2.1 and pool). Run multiple agents at once, each isolated in its own Git worktree. Move sessions between the desktop app, VS Code, and Visual Studio. Download models directly from Hugging Face and work fully offline. Whatever, be creative. Get those agents working for hours and push past old notions of AI We built it because we wanted and needed it ourselves as Laguna pushed what agents were capable of doing. Now it’s yours. Have fun poolside.ai/blog/introducing…

We actually do have an answer to this - more soon Thanks for the push @nathanbenaich

Laguna XS 2.1 already runs on a Mac. Now it's time to make it fly. We teamed up with @eigenlabs on MLX.fast - an open autoresearch competition to optimize Laguna XS 2.1 inference on Apple Silicon. Point your agent at it and improve the open weight ecosystem together. If you think in tokens/sec and low-level perf, this one is for you! mlx.fast/

At 6% the size of DeepSeek V4 Pro, Laguna S 2.1 delivers significantly higher performance. Today, we’re teaming up with @poolsideai to make Laguna XS 2.1 (same architecture as Laguna S 2.1) even faster on consumer Macs. We’re launching MLX(.)fast: a public competition where anyone can collaborate, contribute optimizations, and accelerate Laguna’s MLX inference engine.

Excited to launch MLX.fast with @eigenlabs today. It's an open autoresearch competition to make Laguna XS 2.1 inference as fast as humanly (and agentically) possible on consumer Macs. Eigen's agents already found 36.8% faster inference, and that's before the competition even started. The best part of open weights is that the community takes a model further than any of us could. 
Can't wait to see what everyone does on the leaderboard!  mlx.fast/