Poolside

[San Francisco, California, USA][founded 2023][backed by Bain Capital Ventures; NVIDIA]poolside.ai

Founded in May 2023 by former GitHub CTO Jason Warner and Eiso Kant, Poolside builds open-weight foundation models for agentic coding. The company trains its Laguna model family from scratch on clusters of thousands of NVIDIA GPUs, focusing on long-horizon software engineering tasks. Laguna models consistently outperform competitors several times their size on benchmarks like SWE-Bench and Terminal-Bench.

Models by Poolside (2)

Laguna S 2.1

Poolside

118B MoE (8B active) coding model with 1M context, strong agentic performance, and open weights under OpenMDW-1.1

Context Length1,048,576

Release Date2026-07-21

[text][open-weights][3d teardown]

Show Workflow (1)

Laguna XS 2.1

Poolside

Compact 33B agentic coding model, FP8 quantised for fast inference with 256K context

Context Length262,144

Release Date2026-07-02

[text][open-weights][3d teardown]

Show Workflow (1)