OpenAI has introduced GPT-5.4, its newest flagship artificial intelligence model designed to improve reasoning, coding performance, and the ability to interact directly with computer systems. The release also includes two specialized variants called GPT-5.4 Thinking and GPT-5.4 Pro, expanding the company’s lineup of advanced models for both everyday users and enterprise customers.
The launch signals another step in OpenAI’s strategy to build AI systems that do more than generate text. With the addition of native computer-use abilities and a massive context window, the new model is aimed at handling complex professional tasks that involve multiple tools, documents, and workflows.
GPT-5.4 builds on capabilities introduced in earlier versions while integrating improvements from OpenAI’s coding systems. In particular, the model merges features from GPT-5.3 Codex with stronger performance on everyday professional tasks such as spreadsheets, presentations, document analysis, and structured data work.
The result is a system that can operate across multiple types of work rather than focusing on a single domain. Analysts, developers, researchers, and business professionals are among the primary audiences for the upgrade.
OpenAI says the model is designed to help users complete complex projects that involve multiple steps and different tools. Instead of generating one-off responses, GPT-5.4 aims to support longer and more structured workflows.
One of the most notable additions is GPT-5.4 Thinking, a variant that displays its reasoning process while generating answers. This approach allows users to observe how the model reaches conclusions and adjust instructions during the process.
By showing intermediate reasoning steps, the system gives users more control over complex tasks. If the AI appears to be heading in the wrong direction, users can intervene before the final result is produced.
This feature is designed to make AI assistance more transparent and easier to guide, particularly for technical work or research where the reasoning path matters as much as the final answer.
Another major change is the addition of native computer-use functionality through the API. This capability allows AI systems built on GPT-5.4 to operate software tools and applications directly.
Developers can build AI agents that open programs, manage files, navigate websites, or coordinate tasks across multiple applications. The model can effectively interact with computer environments rather than simply responding with text.
This capability opens the door for more advanced automation, where AI systems act as digital assistants that manage workflows across spreadsheets, documents, code editors, and web tools.
OpenAI says GPT-5.4 supports a 1 million token context window, one of the largest available in a commercial AI model. This allows the system to process extremely large amounts of information at once.
A larger context window means users can feed the model long documents, entire datasets, or extensive codebases without splitting them into smaller segments. This is particularly useful for tasks like legal analysis, financial modelling, or technical documentation.
The expanded capacity also makes the model more capable of maintaining context over long conversations or complex projects.
According to OpenAI, GPT-5.4 delivers stronger performance while using fewer tokens than its predecessor GPT-5.2. That improvement translates into faster responses and lower costs for many tasks.
The company says the model can reach similar conclusions using less computation, making it more efficient for both ChatGPT users and developers using the API.
For coding tasks specifically, GPT-5.4 can produce results with up to 1.5 times faster token velocity, which refers to how quickly the model generates useful output.
OpenAI reports that GPT-5.4 is its most factual model to date. Internal testing indicates that individual statements are roughly 33 percent less likely to contain false information compared with GPT-5.2.
Across full responses, the model produces 18 percent fewer errors than the previous version.
Performance gains are also visible in professional benchmarks designed to evaluate knowledge-work quality. On the GDPval benchmark, which measures performance across 44 occupations, GPT-5.4 scored 83 percent, compared with 70.9 percent for GPT-5.2.
In internal spreadsheet modelling tests, GPT-5.4 achieved 87.3 percent accuracy, a substantial improvement over the 68.4 percent score recorded by GPT-5.2.
The model also performs strongly on computer-use benchmarks such as OSWorld-Verified and WebArena Verified, which test an AI system’s ability to interact with software environments.
Early users in the technology and finance sectors have highlighted the model’s ability to produce complex deliverables.
Mercor CEO Brendan Foody said the system performs well on tasks that require sustained reasoning and structured output. According to Foody, GPT-5.4 is particularly effective at generating long-form materials such as slide decks, financial models, and legal analysis.
He also noted that the model operates faster and at lower cost compared with some competing frontier AI systems.
OpenAI has rolled out GPT-5.4 across both its consumer and developer platforms.
Within ChatGPT, the GPT-5.4 Thinking model is available to Plus, Team, and Pro subscribers. Meanwhile, GPT-5.4 Pro is available to Pro and Enterprise users, offering expanded capabilities for high-end use cases.
OpenAI said the older GPT-5.2 Thinking model will remain accessible for three months before being retired on June 5, 2026 for paid users.
For developers, GPT-5.4 is accessible through OpenAI’s API with the following pricing structure:
The different pricing tiers are designed to support both everyday AI applications and advanced enterprise systems that require deeper reasoning capabilities.
The introduction of computer-use features and large context windows reflects a broader shift in how AI models are evolving. Instead of serving primarily as chat interfaces, new models are increasingly designed to act as autonomous assistants capable of completing multi-step tasks.
GPT-5.4 represents another step in that direction. By combining reasoning transparency, improved efficiency, and the ability to interact with software tools, the model moves closer to functioning as a general digital collaborator.
As AI continues to expand into professional workflows, tools like GPT-5.4 may play a growing role in how individuals and organizations manage complex information and automate everyday tasks.
Be the first to post comment!
