Sakana Fugu adds Nemotron to prove open models can beat GPT-4 class
sakana ai integrated nvidia's open nemotron models into its fugu orchestrator to mix specialist agents for better performance without vendor lock-in.
sakana ai updated its fugu orchestrator to support nvidia's nemotron models, specifically using the 550b-parameter nemotron 3 ultra and the multimodal nemotron 3 nano omni for tasks like coding and tool use. fugu dynamically routes queries between these open models behind a single api, claiming performance on par with proprietary systems while reducing dependence on single vendors. this move signals that the next phase of ai scaling may rely on routing and orchestration rather than training larger monolithic models.