July 21, 2026
Small model, huge main character energy
Laguna S 2.1
Tiny-ish AI drops big scores and commenters are losing their minds
TLDR: Poolside says its new Laguna S 2.1 coding AI performs far above its size, raising hopes for a powerful tool that regular users can actually run. Commenters swung from shock to hype, with many calling it the long-missing middle ground between expensive giants and weaker small models.
Poolside just unveiled Laguna S 2.1, a new coding AI that the company says can go toe-to-toe with much larger rivals while being small enough to run in more realistic setups. In plain English: this is being pitched as a surprisingly compact model that still performs like one of the big kids. Poolside also tried to win trust points by publishing full test run records at trajectories.poolside.ai, which gave the launch an extra dose of receipts or it didn’t happen energy.
But the real fireworks were in the comments, where people reacted like they’d just seen a budget hatchback beat a supercar. One user called it “INSANE,” another said the performance at this size was “quite crazy,” and several zeroed in on the dream scenario: finally, a model ordinary enthusiasts might actually run on personal machines instead of giant corporate servers. That sparked a mini celebration from the self-hosting crowd, with excited chatter about devices like Framework Desktop and Strix Halo and a very specific flavor of nerd joy over downloadable versions already being available.
The hottest take? This is “exactly the kind of model that’s been needed in the middle”: not the absolute king of every leaderboard, but maybe the sweet spot between affordability and usefulness. The underlying drama is classic AI launch season: people are impressed, a little suspicious, and deeply ready to crown a new fan favorite if the real-world results hold up.
Key Points
- •Poolside released Laguna S 2.1, a 118B Mixture-of-Experts model with 8B active parameters per token and up to a 1M-token context window.
- •The company says Laguna S 2.1 went from the start of training to launch in under nine weeks.
- •Poolside reports benchmark results for Laguna S 2.1 across Terminal-Bench 2.1, SWE-Bench Multilingual, SWE-Bench Pro, DeepSWE, SWE Atlas, and Toolathlon Verified.
- •The article states that full trajectories for every trial in the final evaluation set are being published at trajectories.poolside.ai.
- •Poolside highlights a 70.2% Terminal-Bench 2.1 score with thinking enabled and says the model is suited to complex work on local machines due to its compact size.