GPT-5 is a multimodal large language model developed by OpenAI and the fifth in its series of generative pre-trained transformer (GPT) foundation models. Preceded in the series by GPT-4, it was launched on August 7, 2025. It is publicly accessible to users of the chatbot products ChatGPT and Microsoft Copilot as well as to developers through the OpenAI API.
GPT-5 represents a significant advancement in the field of Artificial intelligence, building on the foundations of Machine learning and Deep learning. Its architecture incorporates a Transformer (architecture)-based design, consistent with its predecessors, and it is natively multimodal, trained from scratch on both text and images. The model is designed to handle a wide range of tasks, from complex reasoning to everyday conversational interactions, with a focus on safety and reduced hallucination.
Background
The development of GPT-5 was a long and iterative process. On April 14, 2023, Sam Altman, the chief executive officer of OpenAI, stated at an event at the Massachusetts Institute of Technology that the company was not training GPT-5 at that time, emphasizing a focus on GPT-4. However, by July 18, 2023, OpenAI filed for a "GPT-5" trademark in the United States, and on November 13, 2023, Altman confirmed to the Financial Times that the company was working on GPT-5.
According to The Information, for much of the second half of 2024, OpenAI was developing a model known internally as Orion, intended to become GPT-5. However, the Orion effort failed to produce a better model, leading to its release as GPT-4.5 in February 2025. By late July 2025, anticipation for GPT-5 was high, with reports that Microsoft was preparing for its integration into Copilot. On August 5, 2025, OpenAI released GPT-OSS, a set of two open-weight models with reasoning capabilities, as a precursor. GPT-5 was finally unveiled during a livestream event on August 7, 2025.
Capabilities
At the time of its release, GPT-5 achieved state-of-the-art performance on benchmarks testing mathematics, programming, finance, and multimodal understanding. According to OpenAI, improvements over its predecessors include faster response times, better coding and writing skills, more accurate answers to health questions, and lower levels of hallucination. The model also aims to provide safe, high-level responses to potentially harmful queries rather than outright declining them, an approach called "safe completions," which reduces unnecessary rejections for harmless information. Additionally, GPT-5 was trained to give more critical, less effusively agreeable answers.
Early testers, days before launch, were impressed by its coding and problem-solving abilities, noting significant improvement from GPT-4, though not as large a leap as from GPT-3 to GPT-4. Sam Altman called GPT-5 "a significant step along the path to AGI," referring to artificial general intelligence, and described it as offering "PhD-level" abilities across a wide range of tasks.
The exact energy consumption of GPT-5 has not been disclosed, but researchers at the University of Rhode Island estimated that a medium-length response consumes slightly over 18 watt-hours, equivalent to using an incandescent bulb for 18 minutes.
Architecture
GPT-5 is a system comprising a fast, high-throughput model, a deeper reasoning model, and a real-time router that decides which model to use based on conversation type, complexity, tool needs, and explicit user intent. This design addresses the complexity of manual model selection, which Altman had criticized. The system includes agentic functionality, allowing it to set up its own desktop and use a browser to search autonomously for sources.
The GPT-5 system card defines two fast models (gpt-5-main and gpt-5-main-mini) and two thinking models (gpt-5-thinking and gpt-5-thinking-mini). The API also offers gpt-5-thinking-nano, a smaller and faster version. Developers can adjust reasoning effort (low, medium, high, or minimal) and verbosity (low, medium, or high). ChatGPT provides access to gpt-5-thinking with a setting for parallel test-time compute, referred to as gpt-5-thinking-pro.
Training
GPT-5 is natively multimodal, trained from scratch on multiple modalities (text and images) simultaneously, without relying on pre-trained language or vision models. Its training process involved three stages: unsupervised pretraining, supervised fine-tuning, and reinforcement learning from human feedback (RLHF). Pretraining used a large-scale multilingual dataset of books, articles, web pages, academic papers, and licensed sources. Unlike GPT-4, GPT-5's visual and text capabilities were developed alongside each other throughout training.
Safety and Limitations
Despite its advancements, GPT-5 has faced security concerns. Neuraltrust, a security research company, claimed to have compromised GPT-5 within its first day of testing, enabling it to generate detailed instructions for manufacturing explosive devices. SPLX, another company, conducted similar tests and reached similar conclusions. These assessments suggest significant security gaps, potentially making GPT-5 unsafe for corporate environments.
Use and Integration
GPT-5 is integrated into ChatGPT, where it is free for all users, with higher usage limits for Plus subscribers and unlimited access for Pro users. The introduction of GPT-5 replaced "Advanced Voice Mode" with "ChatGPT Voice," which enables more natural conversations, with Standard Voice Mode retiring on September 9, 2025. On November 24, 2025, a shopping research feature was added, powered by a mini model post-trained on gpt-5-thinking-mini.
GPT-5 is also available in Microsoft Copilot, and Microsoft plans to incorporate it into a wide range of products. According to 9to5Mac, Apple Inc. is planning to integrate the model into Apple Intelligence in iOS 26, iPadOS 26, and macOS Tahoe. Developers can access GPT-5 via the OpenAI API.
Reception and Impact
The release of GPT-5 was met with widespread attention, reflecting its potential to advance Generative AI capabilities. Its performance on benchmarks and its integration into major platforms like Microsoft Azure and Google Cloud (through partnerships) underscore its influence. However, the security concerns raised by Neuraltrust and SPLX highlight ongoing challenges in ensuring the safe deployment of powerful Large language models.
Future Directions
As of late 2025, OpenAI continues to iterate on GPT-5, with plans for further enhancements and broader integration. The model's architecture and training methods are likely to influence future developments in Neural network research and Deep learning applications. The balance between capability and safety remains a critical focus for the industry.