DeepSeek-R1
DeepSeek's reasoning model, which thinks step by step before answering.
Overview
DeepSeek-R1 is DeepSeek's open-weight reasoning model. It works through problems step by step before answering, which makes it strong at maths, coding and logic. Its mixture-of-experts design activates 37B of its 671B parameters per token, so it needs multi-GPU hardware to self-host, and many teams use it through a hosted provider instead.
| Weights | Open |
|---|---|
| Parameters | 671B total, 37B active (mixture of experts) |
| Context window | 128K tokens |
| Licence | MIT |
| Commercial use | Yes |
LLM hosts that offer DeepSeek-R1
All hosts →| LLM host | Hosting plan | Hosting type | Pricing |
|---|---|---|---|
| Amazon AWS | Amazon EC2 | GPU cloud | From US$0.526Per GPU-hour, Reserved or committed capacity |
| Amazon AWS | Amazon Bedrock | Serverless API | From US$0.04Pay per token, Reserved or committed capacity |
| Amazon AWS | Amazon SageMaker | Dedicated endpoint | From US$0.736Per GPU-hour, Reserved or committed capacity |
| Google Cloud | Model Garden | Serverless API | From US$0.07Pay per token, Per GPU-hour |
| Google Cloud | Google Kubernetes Engine (GKE) | GPU cloud | From US$0.35Per GPU-hour, Reserved or committed capacity |
| Google Cloud | Compute Engine | GPU cloud | From US$0.35Per GPU-hour, Reserved or committed capacity |
| Microsoft Azure | Azure Machine Learning | Dedicated endpoint | From US$0.526Per GPU-hour, Reserved or committed capacity |
| Microsoft Azure | Azure Kubernetes Service (AKS) | GPU cloud | From US$0.526Per GPU-hour, Reserved or committed capacity |
| Microsoft Azure | Azure GPU Virtual Machines | GPU cloud | From US$0.526Per GPU-hour, Reserved or committed capacity |