OpenAI Open Weight Models Launch on AWS Expands AI Deployment Options

AWS launches OpenAI open weight models on Amazon Bedrock and SageMaker JumpStart, enhancing AI capabilities for users.

Key Points

  • • New OpenAI models gpt-oss-120b and gpt-oss-20b now available on AWS.
  • • Models support a 128K context window and adjustable reasoning levels.
  • • Available in specific AWS regions: US West for Bedrock and US East, Asia Pacific for SageMaker.
  • • Chain-of-thought output capabilities enhance reasoning interpretability.

In a significant announcement, AWS has made OpenAI's open weight models available on its platforms, Amazon Bedrock and SageMaker JumpStart. This release includes models such as gpt-oss-120b and gpt-oss-20b, which are designed for advanced text generation and reasoning tasks, enabling users to develop AI applications with enhanced capabilities and flexible deployment options.

Both models support a notable 128K context window and offer adjustable reasoning levels, allowing developers to tailor the models to their specific needs. Areas of application include coding assistance, scientific analysis, and complex mathematical reasoning, with performance that competes with other leading models.

AWS's Amazon Bedrock allows for seamless experimentation with these models, enabling users to integrate various functionalities without needing to rewrite existing code. Additionally, production deployment can be achieved through Amazon Bedrock AgentCore, providing a fully managed serverless runtime. In contrast, SageMaker JumpStart offers an intuitive console for quick evaluation and customization, supporting both individual and organizational users.

Currently, the new OpenAI models are available in particular regions: the US West (Oregon) for Bedrock, and US East (Ohio, N. Virginia) and Asia Pacific (Mumbai, Tokyo) for SageMaker. Notably, the models feature chain-of-thought output capabilities, enhancing transparency in reasoning processes, while ensuring safety and retaining compatibility with the standard GPT-4 tokenizer, allowing for greater flexibility in workflows. For detailed pricing, users can check the respective pages for Amazon Bedrock and SageMaker AI.