AWS Rolls Out DeepSeek-R1 as Fully Managed Model on Amazon Bedrock

March 10, 2025

Details

  • Amazon Web Services has launched DeepSeek-R1 as a fully managed, serverless AI model on Amazon Bedrock, becoming available on March 10, 2025, with advanced reasoning capabilities for enterprise use.
  • DeepSeek-R1, developed by DeepSeek, is built using a mixture-of-experts architecture, is MIT licensed, and scores 79.28% on the AIME math benchmark and 49.2% on SWE-bench for software engineering tasks.
  • The model operates at 75–90% lower inference costs compared to competitors, while supporting accurate coding, data analysis, and multilingual tasks across more than 30 languages.
  • This addition broadens AWS’s third-party AI model portfolio, joining offerings from Anthropic and Amazon Titan to intensify competition in the enterprise AI market.
  • AWS supports cross-region inference in three geographical areas and equips DeepSeek-R1 deployments with enterprise security features, including Bedrock Guardrails for content moderation and encryption to meet compliance needs.

Impact

With DeepSeek-R1, AWS fortifies its presence in the enterprise AI space, targeting organizations that demand both cost savings and robust security. The move directly challenges Google's and Microsoft's cloud AI services, offering regulated industries like finance an alternative with strong compliance safeguards. Early adoption by firms such as Moody's and PwC highlights its potential market impact.

Read more