Poolside
@poolsideai
We build models for agentic coding and long-horizon tasks. Try Laguna:
2 Following 14.7K Followers
Agentic evals are messy. A benchmark score tells you something about model performance, but it also reflects the whole system around it: the harness, sandbox, dependencies, timeouts and sometimes a loophole the agent found in the task.
That’s why trajectories matter so much to us. They show what the agent actually did and whether the score means what we think it does. We publish them to make that evidence transparent and auditable, giving the wider community more to learn from.
Watch @aalSonOfRavi and @ConnorBAdams go deep on all of this with @petergostev from @arena, including some surprisingly creative reward hacks!
Show more
Come hear @mgalle and @sudip_r0y get into the weeds on post-training, agent RL, continuous learning, and what it takes to make models work reliably in production!
First episode of Field Notes, our new monthly series on the leaders shaping AI.
@sudip_r0y, Adaption Co-founder, and @mgalle, post-training lead at @poolsideai discuss the last 5% of reliability and what it actually costs.
Show more
we knew the community could make Laguna faster. 2.6x faster is pretty fun :)
huge congrats to the winner, and to all 35 solvers who spent the last few weeks pushing Laguna XS 2.1 further!
more reasons to build in the open, together 🤝
Show more
Our SENPAI agent is currently no.1 on the @poolsideai x @eigenlabs inference optimization comp (for now) and first to break 200 TPS decode
Lovely first validation after a complete re-write of the agent to use @OpenHandsDev instead of cc
Show more
Poolside are pioneering models built specifically for local hardware.
Laguna S 2.1 is a great model for DGX Spark / MacBook. The number of tokens generated is probably an order of magnitude more if you include tokens generated locally.
Show more
One of our engineers asked @poolsideai's Laguna S 2.1 to transform a 715-file C++ game from neon cyberpunk into an Ancient Greek aesthetic.
It orchestrated 3 different models, refactored the code, and produced a playable build.
Show more
Made some improvements to bcode + local models
Biggest winner is Laguna S2.1 which was compromised by provider issues, it gained +23% score since last post, now is very competitive
Show more
excited to share we have made Laguna 2x faster on consumer Mac machines!!
i want to thank all the participants of this challenge -- this wouldn't have been possible without you.
all this with no speculative decoding; we're going to introduce it soon -- we want to make sure we ship an anti-hack verifier for our system!
Show more
Building AI in the open makes the world more secure.
We’re joining the Open Secure AI Alliance to build open tools that safeguard software and agents.
We’ll keep releasing model weights and evaluations and sharing our research to strengthen a broader open ecosystem for defenders.
Excited to contribute alongside @nvidia and the other organizations building this ecosystem.
Show more
Attackers have frontier AI. Defenders need a frontier AI ecosystem—the best open and closed models, force-multiplied by a global community.
During the Hugging Face incident, closed AI blocked essential forensics. An open-weight frontier model helped contain the intrusion.
That’s why we created the Open Secure AI Alliance.
Show more
Today we're releasing Laguna S 2.1, our most capable model to date.
It's a 118B total parameter Mixture-of-Experts model with 8B activated per token, a context window of up to 1M tokens, and thinking and no-thinking modes.
Capable enough to hold its own against models many times its size. Small enough to run on a single @NVIDIAAI DGX Spark.
Laguna S 2.1 is fully open under OpenMDW-1.1, with weights available today on @huggingface
Show more
Turns out, the next Laguna also fits on a @NVIDIAAI DGX Spark 👀
Today we’re releasing Laguna XS 2.1.
It’s a small upgrade to the Laguna XS.2 model, the same 33B total / 3B active MoE and stronger results on multilingual coding and terminal-style tasks.
Available now on @huggingface, @OpenRouter, and via Poolside API.
Show more
setup hell kills good RL ideas.
so we’re giving researchers Laguna XS.2, @PrimeIntellect Lab, and a weekend in London to run the whole loop:
tasks → evals → rewards → training → rollouts → adapters → inference
14 days to go.
come touch the weights:
Show more
Poolside is hosting a 2-day model research hackathon in London.
Join us to push an open-weight agent model as far as you can. RL and fine-tune Laguna XS.2, our latest-generation model, on Prime Intellect Lab.
Dates: May 29–30
Partners: @nvidia + @PrimeIntellect + @huggingface
Prize: NVIDIA DGX Spark
Agents need better models.
Better models need cracked researchers.
Link below.
Show more
