

Poolside
β Text source-checked Nov 22, 2026Poolside builds open-weight foundation models and the systems that refine and improve them.
Source-checked against the vendor's site Β· rebuilt from the vendor's own pagesProfile updated Link checked How we rate
Poolside is a company building open-weight foundation models and the systems that refine and improve them. The Laguna s 2.1 model offers frontier-class reasoning at mid-size cost, optimized for quality, speed, and efficiency, with 118B total parameters, 8B active parameters, and a 1M context window. The Laguna xs 2.1 model is described as the lightest and fastest agentic coding model, small enough to run on-device, with 33B total parameters, 3B active parameters, and a 256K context window.
The model weights are available for download on Hugging Face. The research behind the models covers topics including GPU-to-GPU weight transfers, automated architecture ablations, and reinforcement learning from code execution at scale.
Poolside offers a specific pathway for government missions to bring frontier AI to their operations.
According to Poolside: Poolside is a company building open-weight foundation models and the systems that refine and improve them. Its pages list this strength: Laguna s 2.1 provides frontier-class reasoning at mid-size cost while being optimized for quality, Speed, and efficiency.
β The AI Tool Bible editorial team
Pros
- β Laguna s 2.1 provides frontier-class reasoning at mid-size cost while being optimized for quality, speed, and efficiency.
- β Laguna xs 2.1 is the lightest and fastest agentic coding model, small enough to run on-device.
- β The model weights are available on Hugging Face.
Facts
- Pricing
- enterprise (not source-checked)
- API available
- Yes (not source-checked)
- Open source
- Yes (not source-checked)
- Model used
- Poolside Laguna 2.1 (in-house) β Laguna S 118B/8B-active MoE and Laguna XS 33B/3B-active MoE (not source-checked)
- Hacker News
- 17 stories Β· top 416 points Β· last Aug 18, 2026 Β· checked Nov 18, 2026
- Earliest Wayback capture
- 2022-04-13 (earliest archived capture of the site, not a launch date)