Overview / Description
Laguna by Poolside is an AI developer tool: a family of foundation models built for agentic coding and long-horizon software work. Released by Poolside, the family includes Laguna M.1 (a 225B-parameter mixture-of-experts model with 23B active parameters) and Laguna XS.2 (a 33B model with 3B active parameters published as open weights under the Apache 2.0 license). XS.2 is small enough to run on a single GPU yet, per Poolside, holds up against models several times its size. Both models support context lengths up to 256K tokens, making them suited to multi-file, multi-step coding tasks rather than single-prompt completions. Laguna is distributed through several channels: Poolside's own API, OpenRouter, Ollama, and Hugging Face, so teams can self-host the open-weight XS.2 or call the larger M.1 hosted. The models also power Poolside's own products — pool, a terminal-based coding agent, and Shimmer, a cloud development experience for iterating on web apps, APIs, and CLIs. Laguna is aimed at individual developers, small teams, startups, universities, and research institutions that want strong agentic coding models with an open-weight option. For buyers comparing the best AI for agentic coding, Laguna's draw is the Apache-2.0 open-weight tier plus a long 256K context window.
Used For
Agentic coding and long-horizon software development for developers, startups, and research teams, with an open-weight self-hostable option
Pricing
Plan
Laguna XS.2 available as open weights (Apache 2.0) to self-host at no license cost
Pros & Cons
Pros
- Laguna XS.2 (33B, 3B active) ships as open weights under the Apache 2.0 license
- XS.2 is small enough to run on a single GPU
- Supports context lengths up to 256K tokens for multi-file, long-horizon coding
- Available via Poolside API, OpenRouter, Ollama, and Hugging Face
- Purpose-built for agentic coding rather than general chat completion
Cons
- Long-term pricing for the hosted API is not yet specified (free for a limited time)
- Running the larger M.1 (225B) model self-hosted needs substantial GPU resources
- Newer model family with a shorter public track record than established coding models
- Best results require an agentic workflow (pool, Shimmer, or compatible tooling)
Questions & Answers
Alternatives
Qwen2.5-Coder, DeepSeek-Coder, Codestral, Claude (Anthropic)
Reviews & Ratings
0 reviews
Sign in to rate and review Laguna by Poolside.
Sign in to reviewNo reviews yet. Be the first to review Laguna by Poolside!