NVIDIA and LangChain claim an open AI agent stack matches closed models at one-tenth the inference cost — type0 | type0