SambaNova is a hosted inference service and hardware platform that runs AI models on its own RDU chips, which use a dataflow architecture. You can use SambaCloud, the hosted option, or run SambaStack on SambaRack systems on-premises.
The API is OpenAI-compatible, it hosts open-weight models such as GLM and MiniMax, and it supports bringing your own checkpoint. SambaOrchestrator handles auto scaling, load balancing, monitoring and model management.
Create an account at cloud.sambanova.ai, point an OpenAI-compatible client at its endpoint, and see docs.sambanova.ai. Pricing is on the plans page.