SambaCloud

SambaCloud is SambaNova's pay-per-token LLM inference API for open models such as DeepSeek, Llama and gpt-oss, with a free tier and enterprise plans.

SambaCloud runs large open models on SambaNova's own chips and serves them through an API, so a team can call DeepSeek, Llama or gpt-oss without hosting a model itself. It is aimed at work where response speed matters, such as agents that chain many model calls together. SambaNova, founded in 2017, designs the chips, builds the systems and operates the service. The product also appears under the name SambaNova Cloud. The console holds a Playground for trying models, along with API Keys, Usage, Billing, and Commits and Credits pages. Requests follow the OpenAI format, so existing OpenAI client libraries can call it. Text generation supports function calling through the tools parameter and JSON schema output through response_format. Thinking mode switches on for Gemma 4 with enable_thinking, and MiniMax models bill cached prompt tokens at a reduced rate. Beyond chat, the service offers a Mistral embeddings model, Whisper for transcription and translation, and image input on a preview Gemma model. The service is built for developers, startups and AI teams that want open models served quickly without running their own inference servers. Agent builders, coding tool makers and teams building retrieval systems form its core audience. It is bought by usage: an account starts on the Free Tier, moves to the Developer Tier and per-token billing once a card is linked, and larger customers agree an Enterprise plan with sales. Companies already on AWS can also buy it through AWS Marketplace and connect over PrivateLink. Typical jobs are multi-step agents, retrieval over company documents, deep research and coding help. An agent team can run CrewAI crews on SambaCloud models, while a developer can point the Cline coding assistant at it from the editor. AI Starter Kits in the docs give working examples, such as an Analyst Assistant that answers questions about uploaded PDFs using the embeddings endpoint. Hugging Face users can also pick SambaNova as an inference provider for supported models. For an open-model application, that puts chat, embeddings and speech behind one endpoint and one bill.