Poolside Returns With Open-Source AI Model to Challenge China

Poolside, the AI coding startup co-founded by former GitHub CTO Jason Warner, has emerged from an 18-month quiet period with a new open-source model it claims can compete with the best from China—almost.

The company's Laguna model outperforms American and Chinese open-source competitors on public benchmarks, according to Warner, with one significant exception: Moonshot's Kimi K3 model from China. The admission underscores how Chinese labs building open-source AI have gained ground while U.S. companies focused on closed, commercial systems.

From $3 billion valuation to radio silence

Poolside raised $500 million at a $3 billion valuation in October 2024 to build coding agents for governments and enterprises. Then the company went dark as OpenAI and Anthropic scaled to near-trillion-dollar valuations and Chinese open-source models from labs like Deepseek and Qwen gained adoption.

Warner told Forbes the startup spent those 18 months building what he calls a "model building factory"—infrastructure designed to produce new iterations every five weeks rather than through what he describes as an "artisanal process." The company is positioning this industrial approach as its competitive advantage in a market increasingly dominated by Chinese open-source alternatives.

Why it matters

The competitive landscape for open-source AI has shifted dramatically toward China while American companies chased closed, commercial models. Poolside's return highlights a strategic gap: U.S. labs now find themselves playing catch-up in open-source development, a category that matters for developers, researchers, and organizations that need transparency or want to avoid vendor lock-in. Warner's acknowledgment that Laguna trails Kimi K3 in the 118-billion-parameter weight class signals that Chinese labs have established technical leadership in at least some segments of open-source AI.

The open-source battleground

Warner frames Poolside's position carefully: "As an American company building for the West, we'll be the most capable open model in the West," he said. "Globally, in this weight class of the 118 billion parameter model, we are the leader"—except, he concedes, for Kimi K3.

That qualifier matters. Chinese open-source models including Kimi, Deepseek, and Qwen have become dominant tools, forcing American startups to compete on both performance and the speed of iteration. Poolside joins other U.S. efforts including Mira Murati's Thinking Machines and Reflection AI in attempting to reclaim ground in open-source AI development.

The company's focus on coding agents for government and enterprise customers may give it a niche where national security concerns favor Western providers, even if Chinese models hold technical advantages. But Warner's emphasis on building a repeatable production system suggests Poolside is betting on velocity and consistency rather than one-time breakthroughs.

Details of Poolside's return and Warner's comments were first reported by Forbes staff writer Iain Martin.