OpenAI Releases GPT-54 AI Models With Agentic Computer-Use Capabilities

OpenAI released the GPT-5, a that was also released by Open AI. Friday, he is paraphrasing 4 artificial intelligence (AI) model. A third major update to the GPT-5 foundational model is bringing improvements across parameters and new agentic capabilities, including computer-use. unlike GPT-5 . A new large-language model (LLM) is being released for both ChatGPT and the application programming interface (API) as well as 3-Codex, which was a single code language model that was only developed to be used as coding models for the Codex platform. A more advanced version of the model dubbed GPT-5 was also released by San Francisco-based AI giant, as well. 4 Pro, .

GPT-5.4 AI Models Released

The AI giant announced and detailed the release of the new AI models in a post, which was published in an article about the launch of these new versions of AI. GP5-5 GPT-5 ? There is 4 in both Codex and API, as well as in ChatGPT (as gpt-5). 4 thinking 4 Thinking. At launch, the model will only be available to subscribers of the Plus, Team and Pro at launch (instead of GPT-5). 2 Thinking thinking. Similarly, the GPT-5 was issued on the other hand. Only Pro and Enterprise subscribers can now download 4 Pro.

The GPT-5 is a , OpenAI says. 4 model also makes the reasoning, coding and agentic workflows of 4 models improved. In addition, it includes 5 coding capabilities of . Similarly, 3-Codex (and other general-purpose behaviour) improves the use of tools and perform tasks such as spreadsheets, presentations or documents.

The most interesting new feature is computer-use, though, is the use of a computer. The term “LLM” is the deployment of AI agents that can access a virtual machine and perform complex tasks such as web browsing, placing orders or more. OpenAI – GPT-5 claims . 4. Writer Code to use computers, issue mouse and keyboard commands; steerable behaviour.

The AI giant said the model scored 75 percent success on the OSWorld-Verified benchmark test, which measures a model’s ability to navigate if it is using screenshots and keyboard and mouse actions, according to internal tests.

This is a better model for computer-use, according to OpenAI, which has improved visual perception capabilities. GPT-5 – an image analysis of an AI model’s visual reasoning, on the MMMU-Pro benchmark (which measures one’Sighting Intelligence Model) and its performance in terms of Visual Reason. 4 81 paraphrasing 2 percent success rate.

In a bid to protect itself, the company said GPT-5. 4 is categorized as High cyber capability, and is deployed with an expanded cyber safety stack that includes monitoring systems, trusted access controls (and asynchronous blocking) and other surveillance. In addition, OpenAI has also rolled out chain-of-thought (CoT) controllability as another open-source assessment. In its analysis, it determines whether models can deliberately obfuscate their reasoning to avoid being monitored for the sake of privacy.