Poolside
Coding-model lab shipping open-weight Laguna models and a run-it-yourself agent platform
Overview
Poolside is a San Francisco lab, started in April 2023 by Jason Warner and Eiso Kant, that trains models for agentic coding. Its current open-weight models are Laguna S 2.1 (118B, July 2026) and Laguna XS 2.1 (33B), released under OpenMDW-1.1, alongside the pool coding agent, a macOS Desktop Assistant and an enterprise Platform deployed inside a customer's own boundary.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Poolside
- Laguna S 2.1: 118B mixture-of-experts (8B active), 1M context, open weights under OpenMDW-1.1
- Laguna XS 2.1: 33B mixture-of-experts (3B active), 256K context, sized for local use
- pool terminal coding agent; Laguna also works in any Agent Client Protocol client
- Poolside Desktop Assistant for macOS plus VS Code and Visual Studio extensions
- Hosted access via the Poolside API, OpenRouter, Vercel AI Gateway and Baseten
- Poolside Platform: agent sessions with recorded trajectories on bare metal, air-gapped Dell hardware or an AWS, Azure or Google Cloud VPC
- Forward-deployed research engineers for enterprise rollouts
Poolside has become one of the more interesting open-weight coding labs. Laguna S 2.1 is a 118B mixture-of-experts model with 8B active parameters and a one-million-token context, its weights are downloadable under OpenMDW-1.1, and Poolside publishes the full evaluation trajectories behind its benchmark claims. The smaller XS 2.1 is built to run on a local machine.
Hosted use is cheap or free for now, through Poolside's API, OpenRouter and Vercel AI Gateway, and the enterprise Platform targets regulated buyers who want agents on their own hardware or VPC. The risk is the company rather than the models: its Project Horizon data-centre partnership with CoreWeave ended in April 2026 and its Shimmer preview shut in June. Because the weights are open, teams can adopt Laguna without betting on Poolside's balance sheet.
What we found
Poolside is a San Francisco foundation-model lab building models for agentic coding. Jason Warner and Eiso Kant started it in April 2023 and run it as co-CEOs; the operating company is Poolside, Inc., a Delaware corporation. For its first two years it sold only to enterprises, becoming a first-party AWS offering through Amazon Bedrock and EC2 in December 2024, two months after announcing a $500 million raise. It bought London-based Fern Labs in November 2025.
2026 turned it into an open-weight lab. In April it announced Laguna M.1 (225B parameters, 23B active) and Laguna XS.2 (33B, 3B active, Apache 2.0), its first public models. Laguna XS 2.1 followed on 2 July and Laguna S 2.1 on 21 July: a 118B mixture-of-experts model with 8B active parameters and a one-million-token context, released under OpenMDW-1.1. Poolside reports 70.2% for S 2.1 on Terminal-Bench 2.1 with maximum thinking and publishes every evaluation trajectory, which makes the scores auditable, though they are its own runs.
The weights are free to download, Poolside's API is "free to use for a limited time", and OpenRouter lists free endpoints for both current models alongside paid ones from $0.06 per million input tokens. The pool terminal agent and a macOS Desktop Assistant, which can also drive Claude Code, Codex or Gemini through the Agent Client Protocol, are the first-party front ends. For companies, the Poolside Platform runs agents on bare metal, on air-gapped Dell hardware or in a customer's own AWS, Azure or Google Cloud VPC, on a subscription Poolside does not price publicly.
The financing side is shakier. In April 2026 CoreWeave and Poolside ended their partnership on Project Horizon, a planned two-gigawatt Texas data centre campus, leaving Poolside looking for other cloud partners and funding. Its Shimmer cloud sandbox closed on 8 June. For ASEAN teams, the draw is open weights they can run on hardware they control.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Poolside's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Poolside unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Poolside directly →
Spotted something out of date? Suggest an update →
More in Code & Dev Tools
