OpenAI has announced the release of two open weight models, gpt-oss-120b and gpt-oss-20b, their first open-weight language models since GPT‑2. Both gpt-oss-120b and gpt-oss-20b can be deployed and fine-tuned in Oracle Cloud Infrastructure (OCI) Data Science, joining many other open-source generative AI models that are supported on the platform. These models are available under the Apache 2.0 license and can surpass similarly sized models on reasoning tasks along with powerful tool use capabilities. 

Highlights of OpenAI’s models

Here are the new OpenAI open weight models: 

  • gpt-oss-120b — designed for production, general-purpose and high-reasoning use cases.  The model has 117B parameters with 5.1B active parameters
  • gpt-oss-20b — designed for lower latency and local or specialized use cases.   The model has 21B parameters with 3.6B active parameters

These models were trained with a mix of reinforcement learning and techniques based on OpenAI’s other internal models. According to OpenAI, their performance are on par or exceed OpenAI’s internal models. Both models perform strongly on tool use, few-shot function calling, CoT reasoning and HealthBench.

In addition, gpt-oss models perform comparably to OpenAI’s frontier models on internal safety benchmarks, offering developers the same safety standards as recent proprietary models. 

The weights for both gpt-oss-120b and gpt-oss-20b are freely available for download on Hugging Face and come natively quantized in MXFP4.  Like many other models available on Hugging Face, OCI Data Science supports the gpt-oss models as well.

Working with OpenAI models in OCI Data Science

OCI Data Science supports a Bring Your Own Container approach for model deployment and jobs, which enables you to deploy and fine tune OpenAI’s gpt-oss-120b and gpt-oss-20b.  The Bring-Your-Own-Container approach requires downloading the model from the host repository, Hugging Face, and creating a Data Science model catalog entry. Next, you would download the vLLM container that support the models and push it to the OCI Registry.

Then, you can deploy the model or run a fine-tuning job with the vLLM container image in the OCI Registry. Once the model is deployed, you’re set to invoke the model with an HTTP endpoint. For more details, please check out our tutorials Deploy OpenAI open-source models and Batch Inferencing guide for sample code and instruction.

Get started with OCI Data Science and OpenAI models today!

OCI Data Science enables you to work with the latest open source generative AI models through a best-in-class platform and empowers you to take advantage of the latest open source technologies to power your generative AI workflows.