SambaNova (United States) is the company behind SambaCloud, a cloud inference platform for large open models. The API is compatible with the OpenAI client and with Anthropic’s Messages API, so existing code can be reused by changing only the key and the base URL.
What it does
- Runs hosted open text, image and audio models through a managed API
- Serves applications, agents and RAG flows from a single endpoint
- Lets teams prototype in a self-service portal and scale usage with demand
How it works
- Developers create an account in the portal, generate a key and point the client at the base URL
- Each account can keep several active API keys
- Request and token limits differ between the free tier and the developer tier
- Billing follows consumption, with commercial terms available for larger volumes
- Prompts are not used to train models, according to the official privacy policy
Models and infrastructure
- Production and preview models from vendors such as Meta, DeepSeek, MiniMax and OpenAI
- Execution on the company’s own RDU hardware, designed for large-model inference
- Text generation, vision, audio, embeddings, function calling and prompt caching
- For dedicated environments, the company offers SambaStack, installed on the customer’s infrastructure
Strengths
- OpenAI- and Anthropic-compatible APIs reduce migration effort
- Self-service portal with a free tier for testing
- Stated privacy: the platform does not collect prompts or user data
Points of attention
- The focus is hosted open models; it is not a router across external providers
- Dedicated corporate features and on-premises installation require the SambaStack product
- Free-tier limits are low, so production use calls for the paid tier