Introducing DeepSeek-R1
DeepSeek-R1, a publicly available model under MIT license, represents a breakthrough in AI reasoning capabilities, delivering exceptional accuracy and nuanced understanding across complex tasks. With strong reasoning capabilities, it has demonstrated leading performance in mathematics with a 79.28% score on AIME 2024 and software engineering with 49.2% on SWE-bench Verified. This powerful foundation model offers businesses a cost-effective path to deploy sophisticated, scalable AI solutions that adapt seamlessly to specific business requirements.

Benefits
Meet DeepSeek
DeepSeek is revolutionizing AI with breakthrough innovations in reasoning and comprehension. DeepSeek-R1 achieves advanced levels of accuracy and understanding, marking a transformative leap forward in AI capabilities. Through sophisticated natural language processing, DeepSeek's technology tackles complex challenges with exceptional precision, adapting fluidly to diverse business requirements while delivering consistent results. DeepSeek empowers organizations to build powerful, tailored AI solutions that evolve alongside their ambitions—fundamentally changing how businesses solve their most demanding challenges.
Use cases
Demos
To help you deploy DeepSeek-R1 responsibly in production environments, Amazon Bedrock provides comprehensive and customizable safeguards through Amazon Bedrock Guardrails. This includes sensitive information filtering and customizable security controls—particularly valuable for organizations operating in regulated environments. AWS highly recommends integrating guardrails with your DeepSeek-R1 deployments to add protection for your generative AI applications. Learn how to implement these robust safety protections for DeepSeek models using Amazon Bedrock Guardrails.
Model versions
DeepSeek-R1
DeepSeek's advanced foundation model can process text and is designed for complex reasoning and enterprise applications. It delivers exceptional accuracy and contextual understanding for sophisticated business tasks. The model excels at processing complex requirements, technical documentation, and domain-specific knowledge.
Max input tokens: 128K (Compared to DeepSeek’s first-party solution of 64K)
Languages: English, Chinese
Fine-tuning supported: No
Supported use cases: Code generation, mathematical analysis and problem-solving, scientific computing and research, technical documentation generation, enterprise knowledge management, complex data analysis, intelligent decision support, process automation, and educational content development.