OpenAI’s GPT-OSS Models Now on AWS for Faster, Cheaper AI
Amazon Web Services (AWS) has announced the availability of OpenAI’s two new open-weight foundation models—gpt-oss-120b and gpt-oss-20b—on its cloud services Amazon Bedrock and Amazon SageMaker AI. This announcement is considered a major and historic move in the AI technology sector, as it marks the first time millions of AWS customers will have easy access to such powerful, fast, and cost-effective AI models from OpenAI. This initiative will equip businesses, developers, and researchers with cutting-edge tools capable of handling everything from agentic workflows to scientific and technical analysis, coding, and solving complex mathematical problems.
What is OpenAI’s New Open-Weight Model?
OpenAI has released the gpt-oss-120b and gpt-oss-20b as open-weight models, meaning their internal parameters—known as “weights”—are publicly available. This allows users not only to use these models but also to download, customise, fine-tune, and run them locally (on-premises). While these models are not open source in the strict sense—since their complete source code is not public—they still offer a level of control that enables developers and companies to enhance them according to their requirements.
The gpt-oss-120b model consists of around 120 billion parameters, while gpt-oss-20b contains 20 billion parameters. Both models support modern capabilities such as instruction-following, web search, and code interpretation. They also feature a context window of up to 128,000 tokens, enabling them to efficiently understand and process long and complex documents or conversations.
Availability and Integration on AWS
AWS has integrated these models into Amazon Bedrock and Amazon SageMaker AI. Amazon Bedrock serves as a platform that enables companies to efficiently run AI models without worrying about infrastructure management. Its AgentCore feature allows businesses to securely, scalably, and production-ready deploy the models. Meanwhile, the SageMaker AI platform offers an easy process for pre-training, fine-tuning, and deploying models, enhancing the novelty and capability of AI for organisations.
Performance and Price-Performance Advantage
According to OpenAI, the gpt-oss-120b model is three times faster and cheaper than Google’s Gemini model, and delivers five times better performance than China’s DeepSeek-R1. It is also said to outperform OpenAI’s own o4 model by twofold. With hardware support from partners like Cerebras, the model can seamlessly process at speeds of up to 3,000 tokens per second—bringing significant advances in rapid AI responses and complex reasoning tasks.
Security and Reliability
AWS has paired these models with Guardrails technology, capable of blocking up to 88% of harmful content. This feature ensures that the models remain safe, controlled, and trustworthy in commercial environments. OpenAI has also carried out security audits and risk assessments under its Preparedness Framework to prevent misuse and mitigate vulnerabilities.
OpenAI’s Vision and Collaboration
OpenAI Product Lead Dmitry Pimenov stated that the aim of these open-weight models is to unlock new possibilities for developers and businesses. Whether you are an independent developer or a large enterprise, this technology offers more powerful, affordable, and open AI access. Thanks to AWS’s collaboration, these tools have become more accessible and dependable. OpenAI’s goal is to democratise and enrich AI not only for large corporations but also for small and medium-sized businesses.
AWS’s AI Ecosystem
AWS already delivers generative AI solutions through Amazon Bedrock to major companies such as DoorDash, GoDaddy, Pfizer, Siemens, and PGA TOUR. With the addition of OpenAI, AWS’s AI model portfolio now includes major names like DeepSeek, Meta, and Mistral AI, offering customers a collection of over 100 AI models to choose from according to their needs. This strategy positions AWS as one of the most competitive and comprehensive providers in the cloud AI market.
Key Technical Highlights
- Extensive Context Support for Long Documents: With a 128K token context window, these models can handle lengthy conversations, technical manuals, research papers, and more.
- Open-Weight and Customisable: Licensed under Apache 2.0, these models can be freely downloaded, customised, and used by developers.
- Reduced Latency: Cerebras’s wafer-scale AI infrastructure powers these models at high speed, significantly reducing delays.
- Security and Monitoring: Guardrails and a robust security framework help prevent harmful content and manage risks effectively.
OpenAI’s New Path with Open-Weight Models
Significantly, this release marks a partial shift from OpenAI’s traditionally closed-door approach, embracing more openness with open-weight models. These models offer developers the option to deploy either locally or on the cloud, fostering transparency and innovation in AI modelling. OpenAI has also launched a $500,000 Red Teaming Challenge to encourage researchers to identify security vulnerabilities and misuse risks in the models.
Commercial and Research Benefits
Researchers, governments, and industries can now use these new models for language-specific improvements, regional adaptations, and enhanced AI experiences. For example, OpenAI is partnering with the Swedish government to fine-tune a language-specific model. Such capabilities will pave the way for advanced AI applications tailored to diverse linguistic groups and domains.
Conclusion
The availability of OpenAI’s gpt-oss-120b and gpt-oss-20b models on AWS makes AI technology more transparent, affordable, and customisable. This move marks the beginning of a new chapter in cloud and AI, empowering companies of all sizes, researchers, and developers to harness AI models with greater autonomy and flexibility. Platforms like Amazon Bedrock and SageMaker simplify the adoption of these advancements, while partners such as Cerebras ensure production-grade speed and efficiency.
Also Read
