American Open-Source Labs Think They Can Beat China’s Best AI Startups - Forbes

What the report says

Forbes reports that Poolside, the AI startup cofounded by former GitHub CTO Jason Warner, has re-emerged with a new open-weight coding model called Laguna after a period of lower public visibility. The company had raised $500 million at a $3 billion valuation in 2024 to build coding agents for government and large enterprise customers. Warner told Forbes the company used the intervening period to create a more repeatable “model-building” process and said Poolside aims to release stronger versions on a roughly five-week cycle.

According to Forbes, Poolside says Laguna performs ahead of many U.S. and Chinese open-weight rivals on public benchmarks, though not Moonshot AI’s new Kimi K3. Warner acknowledged that Laguna remains behind the most powerful closed systems from OpenAI and Anthropic and behind Kimi K3, but argued that Poolside’s model is much smaller. Forbes reported that Poolside describes Laguna as a 118 billion-parameter model, while Moonshot has said Kimi K3 has 2.8 trillion parameters.

The launch comes as Chinese open-weight models such as Kimi, DeepSeek and Qwen have gained broad developer and business attention, partly because their weights can be downloaded and adapted rather than accessed only through a vendor-controlled service. Forbes notes that U.S.-linked challengers include Poolside, Mira Murati’s Thinking Machines Lab and Reflection AI, while major companies including OpenAI, Google and Nvidia are also active in open-weight releases.

The competition has policy and commercial stakes. Forbes cites debate in Washington over Chinese AI models and reports that Poolside has also been tied to large infrastructure ambitions, including a West Texas data center project and an investment in British neocloud company Fluidstack.