- The new OpenAI models are executed efficiently in minimum hardware, but they have not been tested independently for workloads
- Models are designed for borders use cases where large -scale infrastructure is not always available
- The Apache 2.0 license can foster a broader experimentation in regions with strict data requirements
OpenAI has launched two open weight models, GPT-ASS-120B and GPT-Oss-20B, positioning them as direct challenges for offerings such as Deepseek-R1 and other large language learning models (LLM) that currently shape the ia ecosystem.
These models are now available on AWS through their Amazon Bedrock and Amazon SageMaker AI platforms.
This marks OpenAi’s entrance to the open weight models segment, a space that has so far been dominated by competitors such as the Mistral and Metral AI.
OpenAI and AWS
The GPT-Oss-20B model is executed in a single 80 GB GPU, while the 20B version is directed to the edge environments with only 16 GB of memory required.
Operai states that both models offer solid reasoning performance, coincide or exceed their O4-MINI model in key reference points.
However, external evaluations are not yet available, leaving real performance in varied workloads open to scrutiny.
What distinguishes these models is not only their size, but also the license.
Low Apache 2.0 launched, they are intended to reduce access barriers and support the development of the broadest, particularly in high security or limited resources environments.
According to Openai, this movement is aligned with its broader mission of making artificial intelligence tools more widely usable in all industries and geographies.
In AWS, the models are integrated into the business infrastructure through Amazon Bedrock Agentcore, which allows the creation of AI agents capable of performing complex workflows.
Operai suggests that these models are suitable for tasks such as coding generation, scientific reasoning and problems solving several steps, especially when adjustable reasoning and thinking chain outputs are required.
Your 128K context window also supports longer interactions, such as document analysis or technical support tasks.
The models are also integrated with the developer tools, support platforms such as VLLM, flame.cpp and hug’s face.
With characteristics such as railings and the next support for the imports of import and knowledge of custom models, OpenAi and AWS are launching this as a base ready for developers to build scalable AI applications.
Even so, the launch is partially strategic, positioning OpenAi as a key player in the open model infrastructure, while attacking its most closely technology to Amazon Web Services, a dominant force in cloud computing.
