Poolside
Verified Oct 16, 2026 ActiveOpen-weight coding models, Laguna s and xs 2.1, with an OpenAI-compatible API and a terminal agent
Listed on EffectHub ·
Pricing
- Pricing model
- Contact sales
No rate card is published. The models page says API access is free for a limited time, and the open weights cost nothing to download; commercial and government terms are handled directly with the company.
Use tool ↗About Poolside
Poolside trains its own foundation models for agentic coding and publishes the weights. The current pair is Laguna s 2.1 at 118 billion parameters with 8 billion active, and Laguna xs 2.1 at 33 billion with 3 billion active, released under licences ranging from OpenMDW to Apache 2.0 and downloadable from Hugging Face. The same models can be called through an OpenAI-compatible chat API, run locally in Docker, or driven from the company's own terminal agent harness, a desktop app, or third parties such as OpenRouter and the Vercel AI Gateway. The site says API use is free for a limited time and quotes no rates, and there is a separate government track alongside the commercial one.
- Platforms
- Web · API
- API
- Yes
- Last verified
- Oct 16, 2026
Verified sources
- Open-weight positioning, Laguna s and xs 2.1, and access via OpenRouter, Vercel AI Gateway, desktop and CLIpoolside.ai ↗
- Model parameter counts, licence range, Hugging Face weights, OpenAI-compatible API, Docker inference and the free-for-now notepoolside.ai ↗
Is Poolside legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Domain registered
- Jul 7, 2020 (6 years ago)Registry record (RDAP), registrar NameCheap, Inc., looked up Sep 12, 2026
- TLS certificate
- Issued by Let's Encrypt, valid until Jan 12, 2027Seen in our own connection to the site on Oct 16, 2026
- Last checked
- Oct 16, 2026The site answered
- Operated by
- PoolsideOur research, card verified Aug 23, 2026
Pros & cons of Poolside
One of the few labs training coding-specific frontier models and then publishing the weights, which means you can run Laguna locally, call it as an API, or point an existing agent harness at it through OpenRouter. The sparse-activation sizes — 8 billion active out of 118 — are aimed squarely at making agentic loops affordable. The gap is commercial: no rate card, a free-for-now API, and a company that publishes research and a government track but not its own address.
Pros
- Open weights on Hugging Face under permissive licences, from OpenMDW to Apache 2.0
- Two sizes, 118B and 33B total with 8B and 3B active, so local inference is realistic
- OpenAI-compatible chat API, so existing clients point at it with a base-URL change
- Local Docker inference plus the company's own terminal agent harness
- Reachable through OpenRouter and the Vercel AI Gateway without a direct account
Cons
- No published pricing at all; the free API window has no stated end date
- The company gives no legal entity, address or country on the pages we could reach
- Coding-focused models, so general assistant work is not what they are tuned for
- Self-hosting a 118B model, even sparsely activated, is not cheap hardware
Best for
- Teams that want a coding model they can self-host rather than rent
- Developers building agent harnesses against an OpenAI-compatible endpoint
- Organisations with data rules that rule out sending code to a hosted provider
Not for
- Buyers who need published, stable per-token pricing to plan against
- Anyone wanting a finished IDE product rather than models and a CLI
Watch out: Free-for-a-limited-time API access is not a price; budget for what it costs after the window, or plan to self-host.
Alternatives to Poolside
Poolside head-to-head
Side-by-side fact sheets against the competitors our research or their makers name — pricing, platforms, API and both verdicts on one page.