SGLang, an open-source AI inference optimization project from UC Berkeley's Ion Stoica lab, has spun out as a new startup called RadixArk at a $400 million valuation. The funding round was led by venture capital firm Accel.
Ying Sheng, a former xAI engineer and key SGLang contributor, left Elon Musk's AI company to become RadixArk's co-founder and CEO. The startup focuses on inference optimization—technology that allows AI models to run faster and more efficiently on the same hardware, creating significant cost savings for companies running AI services.
RadixArk continues developing SGLang as open-source while building Miles, a specialized framework for reinforcement learning. The company has begun charging for hosting services while keeping most tools free. The spin-out follows a similar path taken by vLLM, another inference optimization project now widely used by major tech companies.