OpenAI has announced GPT-5.4, the latest iteration in its generative AI model series, positioning it as one of its most capable systems for professional and enterprise workflows. The model introduces improvements in reasoning, coding, and task automation, along with a significantly expanded context window that allows the system to process far larger volumes of information than previous versions.
The update is being rolled out across ChatGPT, OpenAI’s developer API, and the Codex platform. According to the company, GPT-5.4 is designed to support knowledge-intensive work such as document analysis, spreadsheet tasks, presentation creation, and software development.
OpenAI has introduced the model in multiple variants tailored to different use cases. These include a standard version, GPT-5.4 Thinking which focuses on deeper reasoning and complex problem solving, and GPT-5.4 Pro which is aimed at high-performance enterprise workloads.
The release reflects the rapid evolution of large language models as technology companies compete to build systems that can handle increasingly complex business and productivity tasks.
One of the most notable upgrades in GPT-5.4 is its expanded context window. The model can process up to one million tokens of information in a single prompt, allowing it to analyze long documents, large datasets, and extended conversations without losing context. This capability is expected to be particularly useful for tasks such as legal research, financial modeling, academic writing, and enterprise documentation.
OpenAI says the expanded context window also improves the model’s ability to synthesize information from multiple sources within a single interaction.
Beyond scale, the company has focused on improving reliability. Internal benchmarks cited by OpenAI suggest that GPT-5.4 is about 33 percent less likely to generate incorrect claims compared with earlier versions such as GPT-5.2. The system is also reported to produce around 18 percent fewer factual errors overall.
These improvements address one of the most persistent challenges associated with generative AI systems, often referred to as hallucinations, where models produce plausible but inaccurate information.
The model has also shown stronger performance across several evaluation benchmarks designed to measure real-world knowledge work tasks. In tests focused on computer interaction capabilities and digital environments, GPT-5.4 reportedly achieved high scores in benchmarks such as OSWorld-Verified and WebArena-Verified.
Another evaluation, Mercor’s APEX-Agents benchmark, which measures professional skills across fields including finance and law, also indicated improvements in task performance.
OpenAI said the model performs particularly well when generating long-form outputs such as slide presentations, financial analyses, and structured reports.
Another technical upgrade introduced in GPT-5.4 relates to how the system interacts with external tools and software environments. OpenAI has implemented a feature called Tool Search in its API framework.
Instead of loading definitions for every possible tool in the system prompt, GPT-5.4 can dynamically retrieve relevant tool information when required. The company says this approach helps reduce token usage and improves efficiency, which could make large-scale AI deployments faster and less costly.
In addition to reasoning improvements, GPT-5.4 incorporates enhanced coding capabilities derived from earlier models in the GPT-5 series. These upgrades are expected to support developers in writing, debugging, and reviewing code across multiple programming environments.
The model also expands support for what OpenAI describes as “agentic workflows,” in which AI systems can interact with software interfaces and perform multi-step digital tasks.
According to OpenAI, GPT-5.4 can interpret screenshots of software environments and translate them into actions such as keyboard inputs or mouse commands. This feature is aimed at enabling AI systems to operate more effectively within desktop applications and enterprise software tools.
The company says these capabilities are part of its broader effort to move AI systems beyond text generation and toward task execution within real-world digital environments.
GPT-5.4 is now available to users through ChatGPT’s web and mobile applications, as well as through OpenAI’s developer ecosystem. Subscribers to paid tiers such as Plus, Team, and Pro will have access to the reasoning-focused GPT-5.4 Thinking model, while the GPT-5.4 Pro version is expected to be available primarily through enterprise and API channels.
The launch arrives at a time of intense competition in the generative AI sector. Companies including Google, Anthropic, and several startup AI labs have been developing competing models that aim to improve reasoning, coding, and productivity workflows.
Enterprise adoption has become a key battleground in this race, with organizations increasingly exploring how AI systems can automate knowledge work and streamline business operations.
For OpenAI, the introduction of GPT-5.4 signals an ongoing strategy of iterative upgrades rather than single large leaps in model capability. By focusing on reliability, scalability, and enterprise functionality, the company appears to be positioning its AI models as tools for everyday professional tasks.
As generative AI systems continue to evolve, improvements in reasoning accuracy, context handling, and integration with digital tools are likely to shape how organizations deploy AI across business workflows.
With GPT-5.4, OpenAI is aiming to push its models further into the domain of practical knowledge work, where the ability to analyze information, interact with software, and produce structured outputs could redefine how professionals use AI in daily operations.