OpenAI gpt-oss-120b (Beta)
Pre-General Availability: 2025-09-09
This documentation is in pre-General Availability status and is intended for demonstration and preliminary use only. It may not be specific to the hardware on which you are using the software. Oracle Corporation and its affiliates are not responsible for and expressly disclaim all warranties of any kind with respect to this documentation and will not be responsible for any loss, costs, or damages incurred due to the use of this documentation.
This documentation is not a commitment by Oracle to deliver any material, code, functionality or services. This documentation, and Oracle Pre-GA programs and services are subject to change at any time without notice and, accordingly, should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality for Oracle’s Pre-GA programs and services remains at the sole discretion of Oracle. All release dates or other predictions of future events are subject to change. The future availability of any future Oracle program or service should not be relied on in entering into any license or service agreement with Oracle.
See Oracle Legal Notices.
The openai.gpt-oss-120b is an open-weight, text-only language model designed for powerful reasoning and agentic tasks.
Available in These Regions
- Germany Central (Frankfurt) (on-demand only)
- Japan Central (Osaka) (on-demand only)
- US Midwest (Chicago) (on-demand only)
Key Features
- Model Name in OCI
Generative AI:
openai.gpt-oss-120b - Model Size: 117 billion parameters
- Available On-Demand: Access this model on-demand, through the Console playground or the API.
- Text Mode Only: Input text and get a text output. Images and file inputs such as audio, video, and document files aren't supported.
- Knowledge: Specialized in advanced reasoning and text-based tasks across a wide range of subjects.
- Context Length: 128,000 tokens (maximum prompt + response length is 128,000 tokens for each run). In the playground, the response length is capped at 16,000 tokens for each run.
- Excels at These Use Cases: Because of its training data, this model is especially strong in STEM (science, technology, engineering, and mathematics), coding, and general knowledge. Suitable for high-reasoning, production-level tasks.
- Function Calling: Yes, through the API.
- Has Reasoning: Yes.
- Knowledge Cutoff: June 2024
For key feature details, see the OpenAI gpt-oss documentation.
On-Demand Mode
-
You pay as you go for each inference call when you use the models in the playground or when you call the models through the API.
- Low barrier to start using Generative AI.
- Great for experimenting, proof of concepts, and evaluating the models.
- Available for the pretrained models in regions not listed as (dedicated AI cluster only).
Dynamic Throttling Limit Adjustment for On-Demand Mode
For optimized allocation of resources to tenants and to ensure that tenants receive fair access to the models, OCI Generative AI regularly adjusts the request throttling limit for each active tenancy based on model demand and system capacity. This adjustment depends on the following factors:
- The current maximum throughput supported by the target model.
- Any unused system capacity at the time of adjustment.
- Each tenancy's historical throughput usage and any specified override limits set for that tenancy.
Because of the dynamic throttling limit adjustment, we recommend implementing a back-off strategy, which involves delaying requests after a rejection. Without one, repeated rapid requests can lead to further rejections over time, increased latency, and potential temporary blocking of client by the Generative AI service. By using a back-off strategy, such as an exponential back-off strategy, you can distribute requests more evenly, reduce load, and improve retry success, following industry best practices and enhancing the overall stability and performance of your integration to the service.
The OpenAI gpt-oss-120b (Beta) model is available only in the on-demand mode.
| Model Name | OCI Model Name | Getting Access |
|---|---|---|
| OpenAI gpt-oss-120b (Beta) | openai.gpt-oss-120b |
Contact Oracle Beta Programs |
Release Date
| Model | Beta Release Date | On-Demand Retirement Date | Dedicated Mode Retirement Date |
|---|---|---|---|
openai.gpt-oss-120b |
2025-09-09 | Tentative | This model isn't available for the dedicated mode. |
To learn about OCI Generative AI model deprecation and retirement, see Retiring the Models.
Model Parameters
To change the model responses, you can change the values of the following parameters in the playground or the API.
- Maximum output tokens
-
The maximum number of tokens that you want the model to generate for each response. Estimate four characters per token. Because you're prompting a chat model, the response depends on the prompt and each response doesn't necessarily use up the maximum allocated tokens. The maximum prompt + output length is 128,000 tokens for each run. In the playground, the maximum output tokens is capped at 16,000 tokens for each run.
Tip
For large inputs with difficult problems, set a high value for the maximum output tokens parameter. - Temperature
-
The level of randomness used to generate the output text. Min: 0, Max: 2, Default: 1
Tip
Start with the temperature set to 0 or less than one, and increase the temperature as you regenerate the prompts for a more creative output. High temperatures can introduce hallucinations and factually incorrect information. - Top p
-
A sampling method that controls the cumulative probability of the top tokens to consider for the next token. Assign
pa decimal number between 0 and 1 for the probability. For example, enter 0.75 for the top 75 percent to be considered. Setpto 1 to consider all tokens. Default: 1 - Frequency penalty
-
A penalty that's assigned to a token when that token appears frequently. High penalties encourage fewer repeated tokens and produce a more random output. Set to 0 to disable. Default: 0
- Presence penalty
-
A penalty that's assigned to each token when it appears in the output to encourage generating outputs with tokens that haven't been used. Set to 0 to disable. Default: 0