Poolside is an artificial intelligence startup that builds large language models trained specifically to write, test, and reason about software code, then publishes the resulting models for anyone to download and run. It was founded in 2023 on the bet that the fastest path to advanced AI runs through software — that teaching a model to reason well about code is the shortest route to a model that reasons well about everything else.
Who is behind it
Poolside was co-founded in 2023 by Jason Warner, the former chief technology officer of GitHub, and Eiso Kant, who had previously built developer-focused startups including a code search and analysis company. Poolside operates out of San Francisco and Paris, with a smaller office in London. It raised a $126 million seed round in August 2023, backed by Felicis, Redpoint Ventures, and French telecom billionaire Xavier Niel, followed by a $500 million Series B in October 2024 led by Bain Capital Ventures that valued the company at roughly $3 billion; eBay, Nvidia, and Citi Ventures also invested. That level of funding put this AI startup in the same conversation as far larger labs, even though the models it has published so far are a fraction of the size of the biggest frontier systems from OpenAI, Anthropic, or Google DeepMind.
What it builds
Poolside’s large language models are called Laguna, and the lineup spans a few sizes built for different jobs: a small model designed to run on a single developer’s machine, a larger “flagship” released under the permissive Apache 2.0 license, and, as of July 2026, Laguna S 2.1 — a 118-billion-parameter Mixture-of-Experts model that activates only about 8 billion of those parameters for any given request, which keeps it cheap to run despite its size. Laguna S 2.1 reads up to a million tokens of context at once and is built for agentic coding: not just completing a line of code, but carrying out multi-step programming tasks — navigating a codebase, running tests, fixing what fails — over long stretches with minimal human intervention.
Unlike the flagship models OpenAI and Anthropic sell exclusively through an API, Poolside publishes the full weights of its models on Hugging Face under OpenMDW-1.1, a permissive license for AI model weights maintained by the Linux Foundation. That makes Laguna an open-weight model: anyone can download it, run it on their own hardware, fine-tune it, or build a product on top of it without paying Poolside a licensing fee. Developers who would rather not host it themselves can reach Laguna S 2.1 through hosted providers such as OpenRouter, priced at $0.10 per million input tokens and $0.20 per million output tokens for the full context window (as of July 2026, per OpenRouter’s pricing page), or try it directly, no signup required, at Poolside’s own web chat.
Why it matters
Poolside is a bet that the next edge in AI competition won’t come from one do-everything chatbot, but from models narrowly good at the task that matters most to the AI industry’s own growth: writing software, including the software AI labs use to build their next models. That focus puts Poolside up against both dedicated AI coding assistants and the general-purpose frontier models from OpenAI, Anthropic, and Google that increasingly treat coding as just one of many strengths. Its choice to open its weights also lines it up against a wave of strong open models from Chinese labs such as DeepSeek and Alibaba’s Qwen, which have undercut Western labs on both price and openness — making Poolside one of the few American-founded companies matching that approach instead of keeping its best work behind a paywall.
In the news
Poolside’s latest release made today’s briefs: Poolside Releases Laguna S 2.1, an Open-Weight Coding Model.
Sources: Poolside’s Laguna S 2.1 announcement and Hugging Face model card; TechCrunch; Wikipedia.