computer-use-preview
Version: 2025-03-11
Key capabilities
About this model
CUA in the API does not operate computers or browsers. Applications send to CUA screenshots of a computer along with instructions, and the CUA model responds with the actions for the application to take, such as navigating and clicking the pointer, and entering text.Key model capabilities
Core capabilities- Model Qualities: While CUA is still early and has limitations, it sets new state-of-the-art benchmark results, achieving a 38.1% success rate on OSWorld for full computer use tasks, and 58.1% on WebArena and 87% on WebVoyager for web-based tasks. These results highlight CUA's ability to navigate and operate across diverse environments using a single general action space.
- Safety: CUA has been extensively tested for safety, and implements safeguards across several dimensions. CUA refuses many harmful tasks and illegal or regulated activities, is trained to ask users for confirmation before finalizing tasks with external side effects, and is designed to identify and ignore prompt injections on websites.
Use cases
See Responsible AI for additional considerations for responsible use.Key use cases
The provider has not supplied this information.Out of scope use cases
CUA cannot reliably ensure human-in-the-loop intervention. Developers will need to be systematically aware of, and defend against, situations where the model can be fooled into executing commands that are harmful to the user or the system, such as downloading malware, leaking credentials, or issuing fraudulent financial transactions. Particular attention should be paid to the fact that screenshot inputs are untrusted by nature and may include malicious instructions aimed at the model.Pricing
Pricing is based on a number of factors, including deployment type and tokens used. See pricing details here.Technical specs
The provider has not supplied this information.Training cut-off date
The provider has not supplied this information.Training time
The provider has not supplied this information.Input formats
The provider has not supplied this information.Output formats
The provider has not supplied this information.Supported languages
The provider has not supplied this information.Sample JSON response
The provider has not supplied this information.Model architecture
The provider has not supplied this information.Long context
The provider has not supplied this information.Optimizing model performance
The provider has not supplied this information.Additional assets
The provider has not supplied this information.Training disclosure
Training, testing and validation
The provider has not supplied this information.Distribution
Distribution channels
This model is provided through the Azure OpenAI service.More information
The following documents are applicable:Responsible AI considerations
Safety techniques
Safety is built into our models from the beginning, and reinforced at every step of our development process. In pre-training, we filter out information that we do not want our models to learn from or output, such as hate speech, adult content, sites that primarily aggregate personal information, and spam. In post-training, we align the model's behavior to our policies using techniques such as reinforcement learning with human feedback (RLHF) to improve the accuracy and reliability of the models' responses. computer-use-preview or CUA implements additional safeguards across several dimensions. CUA refuses many harmful tasks and illegal or regulated activities, is trained to ask users for confirmation before finalizing tasks with external side effects, and is designed to identify and ignore prompt injections on websites. Details on CUA model are covered in the Operator System card. Prompts and completions are passed through a default configuration of Azure AI Content Safety classification models to detect and prevent the output of harmful content. Learn more about Azure AI Content Safety . Additional classification models and configuration options are available when you deploy an Azure OpenAI model in production; learn more .Safety evaluations
CUA has been extensively tested for safety, and implements safeguards across several dimensions.Known limitations
While CUA is still early and has limitations, it sets new state-of-the-art benchmark results, achieving a 38.1% success rate on OSWorld for full computer use tasks, and 58.1% on WebArena and 87% on WebVoyager for web-based tasks. CUA cannot reliably ensure human-in-the-loop intervention. Developers will need to be systematically aware of, and defend against, situations where the model can be fooled into executing commands that are harmful to the user or the system, such as downloading malware, leaking credentials, or issuing fraudulent financial transactions. Particular attention should be paid to the fact that screenshot inputs are untrusted by nature and may include malicious instructions aimed at the model. We'll continue to monitor how computer-use-preview is being used and improve the model's safety as we identify new risks.Acceptable use
Acceptable use policy
The provider has not supplied this information.Quality and performance evaluations
Source: OpenAI While CUA is still early and has limitations, it sets new state-of-the-art benchmark results, achieving a 38.1% success rate on OSWorld for full computer use tasks, and 58.1% on WebArena and 87% on WebVoyager for web-based tasks. These results highlight CUA's ability to navigate and operate across diverse environments using a single general action space.Benchmarking methodology
Source: OpenAI The provider has not supplied this information.Public data summary
Source: OpenAI The provider has not supplied this information.Model Specifications
Context Length131072
LicenseCustom
Training DataOctober 2023
Last UpdatedNovember 2025
Input TypeText,Image
Output TypeText
ProviderOpenAI
Languages27 Languages