What is Poolside?

Poolside's Laguna XS2 went from start of pre-training to launch in 5 weeks.

“You looked at Laguna XS2 that we launched it was 5 weeks from the beginning of pre-training to launch.”

Poolside runs 10,000 to 20,000 experiments per month with a small team.

“Far more than 10,000 maybe 10 to 20,000 experiments a month.”

Gains in Laguna S come primarily from behavioral changes (more verification, less assuming, greater persistence) rather than raw intelligence increases, clarifying that behavior tuning is distinct from capability scaling.

“A lot of the gains in Laguna s come not from more intelligence, but more from different behavior, more verification, less taking things for granted, not declaring victory early, and being way more persistent.”

RL training time is Poolside's primary wall-clock bottleneck because batch size constraints prevent scaling via additional GPUs, illustrating a non-obvious ceiling on RL parallelization.

“My biggest wall clock bottleneck right now is RL time, right? And it's just because I can't scale it up further because I can't add more GPUs to it because of that bad size constraint.”

Poolside reported zero meaningful on-call wake-up events for engineers across the entire year, signaling exceptional training pipeline reliability.

“There was no on call events right like completely zero and actually we haven't had a meaningful on call event like to wake up for as far as I recall this entire year.”

Eiso Kant on the training run being anticlimactic compared to surrounding R&D work.

“The training run is not the expensive part. The training run is a very anticlimactic event, right?”

Poolside's model factory enabled an engineer to transition into a productive RL researcher within 6 months, suggesting the factory accelerates research iteration beyond just compute throughput.

“One of the guys on our team who started as an engineer building our agents is a legit reinforcement learning researcher now making real progress. and that happened in the span of like 6 months.”

Laguna S outperforms models two to three times its own size on coding benchmarks.

“We are outperforming models two or three times their size.”

Poolside's internal 'polishing' process adapts trained models to perform well on external harnesses beyond their own, a lightweight adaptation step worth tracking as a productization pattern.

“We internally have been kind of this calling this polishing which is like you've got your model you do a little bit of polishing so that like it's able to work well on other harnesses as it is in your own.”

Hugging Face and Poolside are both receiving significant Nvidia capital backing, flagging them as key open-weight infrastructure bets worth tracking.

“Nvidia's backed hugging face and poolside with throwing like a lot of capital.”