SambaNova
Inference on custom dataflow hardware
Last reviewed Sep 28, 2026
SambaNova builds its own AI accelerator, the Reconfigurable Dataflow Unit (RDU), and runs SambaNova Cloud, an OpenAI-compatible inference API for open models served on that hardware and billed per token.
We're actively tracking prices for SambaNova. Check back soon, or browse other providers with current pricing.
Pros & Cons
Advantages
- Serves models on non-GPU hardware built for inference
- Public per-token rates for every served model
- OpenAI-compatible endpoints
Limitations
- Small model catalog compared with GPU-based inference providers
- Some models are priced above other providers of the same open weights
- No GPU rental; inference only
Key Features
Custom RDU hardware
Models run on SambaNova's own Reconfigurable Dataflow Units rather than GPUs
OpenAI-compatible API
Chat completions for open models such as DeepSeek, Llama, Gemma, MiniMax and gpt-oss
Prompt caching
Cached input tokens billed at a reduced rate on supported models
Commits and credits
Prepaid commitments alongside pay-as-you-go billing
Pricing Options
| Option | Details |
|---|---|
| Per token | Input, cached input, and output tokens billed per 1M |
| Commits | Prepaid usage commitments |
Availability & Support
Support
Documentation and the SambaNova Cloud dashboard
Company
Published by SambaNova to identify the company behind this listing.
- Headquarters
- 🇺🇸 Palo Alto, United States
Getting Started
- 1
Create an account
Sign up on SambaNova Cloud
- 2
Create an API key
Generate a key in the API Keys page
- 3
Call the API
Point an OpenAI-compatible client at api.sambanova.ai
Frequently Asked Questions
Which models does SambaNova serve?
Check the pricing table above for the models SambaNova currently serves and their per-token rates.
How do I get started with SambaNova?
Create an account, Create an API key, Call the API
What are SambaNova's main advantages?
SambaNova's main advantages include: Serves models on non-GPU hardware built for inference, Public per-token rates for every served model, OpenAI-compatible endpoints.
What are SambaNova's limitations?
SambaNova's main limitations include: Small model catalog compared with GPU-based inference providers, Some models are priced above other providers of the same open weights, No GPU rental; inference only.