tools
SambaNova DataScale
SambaNova DataScale is an on-premises or hosted AI system for training and running large models with dataflow processors.
SambaNova DataScale is an integrated hardware and software system for AI training and inference. The DataScale SN40L uses Reconfigurable Dataflow Units, a dataflow architecture, and three-tier memory to run multiple models and switch between them quickly.
It is aimed at enterprises, government agencies, and research organizations. Deployments can be on-premises or in hosted data centers. Hardware pricing is not published on the reviewed pricing page; that page lists SambaNova Cloud token usage instead.
Features
- Runs AI model training and inference
- Uses SN40L Reconfigurable Dataflow Units
- Supports on-premises and hosted data-center deployment
- Runs multiple models on one system
- Switches between models in microseconds
- Uses three-tier memory with SRAM, HBM, and DDR
- Supports custom and chained models
- Runs Red Hat Enterprise Linux
Use cases
- Train and deploy large language models
- Run generative and agentic AI workloads
- Analyze high-resolution computer-vision images
- Support scientific research and AI-for-science workloads
- Deploy private AI infrastructure for enterprises
- Run multiple models from a shared system
Pros
Cons
Latest updates
- Supermicro Will Build and Sell SambaNova-Powered Inference Systems (SN50)
Supermicro will deliver and support inference systems pairing NVIDIA GPUs with SambaNova SN50 RDUs, rolling out in phases.
- MiniMax M3.1 Flash Now Available on SambaCloud (MiniMax M3.1 Flash)
MiniMax M3.1 Flash is now available on SambaCloud.
- MiniMax M3 Prompt Caching on SambaCloud: How It Works (MiniMax M3)
Prompt caching is now available for MiniMax M3 on SambaCloud for shared prefixes of at least 4,096 tokens.
- MiniMax M3 Running Fastest on SambaCloud (MiniMax M3)
MiniMax M3 is available on SambaCloud for developers building long-horizon, long-context agents.
- Prompt Caching on SambaCloud: Faster, Cheaper Inference (MiniMax M2.7)
Prompt caching is now live on SambaCloud, starting with support for MiniMax M2.7.
Capabilities
- Choice of models — “SambaStack switches between multiple frontier-scale models, enabling complex agentic AI workflows to execute end-to-end on one node.” source
- Self-hosted — “SambaRack™ is a state-of-the-art hardware platform that can be set up easily in data centers to run Al inference workloads.” source
- API — “SambaNova provides simple-to-integrate APIs for Al inference, making it easy to onboard applications.” source
- Runs models for you — “SambaNova provides simple-to-integrate APIs for Al inference, making it easy to onboard applications.” source
- Builds agents and workflows — “Build or use pre-built AI agents — all with business-aware intelligence.” source
Get it
Pricing
- Prices checked
- 2026-10-24
MiniMax-M3.1-Flash
- 0.03 USD — Cached Input Tokens
- 0.30 USD — Input (per 1M tokens)
- 1.20 USD — Output (per 1M tokens)
DeepSeek-V3.1
- 3 USD — Input (per 1M tokens)
- 4.50 USD — Output (per 1M tokens)
DeepSeek-V3.2
- 3 USD — Input (per 1M tokens)
- 4.50 USD — Output (per 1M tokens)
gemma-4-31B-it
- 0.38 USD — Input (per 1M tokens)
- 1.15 USD — Output (per 1M tokens)
gpt-oss-120b
- 0.22 USD — Input (per 1M tokens)
- 0.59 USD — Output (per 1M tokens)
Meta-Llama-3.3-70B-Instruct
- 0.60 USD — Input (per 1M tokens)
- 1.20 USD — Output (per 1M tokens)