GPT-OSS on AWS: Powering Open, Cost-Effective Generative AI

In a groundbreaking development for the AI community, OpenAI has released its first open-weight models since GPT-2: the GPT-OSS-120b and GPT-OSS-20b. These models are now available on Amazon Web Services (AWS) through Amazon Bedrock and SageMaker JumpStart, offering developers unprecedented access to powerful AI tools. This move not only democratizes AI development but also provides enterprises with scalable, secure and cost-effective solutions for their AI needs.

What Are GPT-OSS Models?

The GPT-OSS models are large language models (LLMs) designed to perform a variety of tasks, including text generation, reasoning and coding. The GPT-OSS-120b model boasts 120 billion parameters, while the GPT-OSS-20b model has 20 billion parameters. Despite their size, both models are optimized for efficiency and can run on hardware ranging from consumer-grade machines to enterprise-level infrastructure.

Released under the permissive Apache 2.0 license, these models are freely available for download and can be customized for specific use cases. The GPT-OSS-120b model has been shown to achieve near-parity with OpenAI’s proprietary o4-mini model on core reasoning benchmarks, while running efficiently on a single 80 GB GPU. The GPT-OSS-20b model delivers similar results to OpenAI’s o3-mini on common benchmarks and can run on edge devices with just 16 GB of memory, making it ideal for on-device use cases, local inference or rapid iteration without costly infrastructure.

Integration with AWS: Amazon Bedrock and SageMaker JumpStart

By integrating these models into AWS’s ecosystem, developers gain access to a suite of tools and services that enhance the capabilities of GPT-OSS models.

Amazon Bedrock provides a fully managed environment for building and scaling generative AI applications. With GPT-OSS models available through Amazon Bedrock, developers can leverage OpenAI’s models alongside other leading AI models through a single, unified API, allowing them to select the best model for each specific use case without changing application code. This integration gives developers the flexibility to modify and customize the models for their particular business needs while maintaining complete control over their data.

Fig 1: Amazon Bedrock Text Playground interface showing configuration of the GPT-OSS-120B model with a prompt example for AI-powered customer support.

Amazon SageMaker JumpStart offers a comprehensive suite of tools for building, training and deploying machine learning models. Developers can quickly evaluate, compare and customize GPT-OSS models for their use cases. Once customized, these models can be deployed in production using the SageMaker AI console or the SageMaker Python SDK. This streamlined process accelerates the development cycle and simplifies the deployment of AI applications.

Fig 2: Amazon SageMaker AI Studio JumpStart interface highlighting the availability of OpenAI models (GPT-OSS-20b and GPT-OSS-120b) for deployment and experimentation.

Key Features and Benefits

• Extended Context Windows: Both models support a 128K context window, allowing them to handle longer documents and more complex tasks without losing coherence.
• Adjustable Reasoning Levels: Developers can fine-tune the models’ reasoning capabilities by selecting low, medium or high reasoning levels, tailoring the models’ performance to specific requirements.
• Tool Integration: The models support external tools to enhance their capabilities and can be used in agentic workflows, such as those built with frameworks like Strands Agents.
• Enterprise-Grade Security: Amazon Bedrock offers enterprise-grade security features, including Guardrails, which help block up to 88% of harmful content using configurable safeguards. This ensures that AI applications built with GPT-OSS models adhere to security and safety standards.
• Cost Efficiency: The GPT-OSS-120b model is three times more price-efficient than Gemini and five times more efficient than DeepSeek-R1 when used through Amazon Bedrock, making it a cost-effective choice for enterprises.

Real-World Applications

The GPT-OSS models are versatile and can be applied across various industries:

• Healthcare: Analyze medical literature, assist in diagnostics and support personalized patient care.
• Finance: Automate financial reporting, perform risk assessments and provide investment insights.
• Education: Develop intelligent tutoring systems, create personalized learning experiences and assist in content creation.
• Customer Service: Enhance chatbots, automate support tickets and provide real-time assistance to customers.

By integrating GPT-OSS models into their workflows, organizations can streamline operations, improve decision-making and deliver better services to their clients.

Getting Started

To begin using the GPT-OSS models on AWS:

1. Access Amazon Bedrock or SageMaker JumpStart: Log in to your AWS account and navigate to the respective service.
2. Select the GPT-OSS Model: Choose either the GPT-OSS-120b or GPT-OSS-20b model based on your requirements.
3. Deploy the Model: Follow the provided instructions to deploy the model in your environment.
4. Integrate into Your Application: Utilize the unified API to integrate the model into your existing applications.

Detailed documentation and tutorials are available on the AWS website to assist you in the setup process. If you are interested in diving deeper into the world of AI, check out courses like Generative AI Essentials to explore advanced concepts and use cases.

Shaping the Future of Open AI on AWS

The availability of OpenAI’s GPT-OSS models on AWS marks a significant milestone in the evolution of generative AI. By providing open-weight models with robust capabilities and integrating them into AWS’s powerful infrastructure, developers and enterprises are empowered to build innovative AI solutions that are scalable, secure and cost-effective. As AI continues to transform industries, the collaboration between OpenAI and AWS sets the stage for the next generation of intelligent applications.

About CloudThat