Released in March 2026, GPT-5.4 (xhigh) is a flagship large language model developed by OpenAI that represents the highest reasoning tier within the GPT-5.4 series. This version unifies the capabilities of the general-purpose GPT models with the specialized code-generation strengths of the Codex series into a single multimodal system. The "xhigh" designation refers to the model's Extreme Reasoning effort level, which is designed for high-complexity professional work, graduate-level scientific analysis, and long-horizon agentic tasks that require deep internal deliberation.
A defining feature of GPT-5.4 is its native computer-use capability, allowing it to interact directly with software interfaces and desktop environments. It has demonstrated frontier-level performance on benchmarks like OSWorld, achieving a success rate that exceeds human baselines for autonomous navigation and task execution. The model is built to handle massive data structures using a 1.05 million token context window, enabling the analysis of entire software repositories or extensive legal libraries in a single inference pass.
Capabilities and Reasoning
OpenAI introduced several architectural improvements with this release, including a deferred tool loading system and a specialized Tool Search engine to manage large ecosystems of external functions more efficiently. The model's factuality has seen a marked improvement over previous iterations; reports indicate it is 33% less likely to produce false individual claims compared to GPT-5.2. Its knowledge cutoff is documented as August 31, 2025.
For developers and enterprise users, GPT-5.4 (xhigh) supports a steerable thinking process where the model can present an upfront plan before execution. This allows users to intervene and refine the reasoning path mid-response. While more verbose and computationally intensive than standard modes, the xhigh configuration provides the necessary depth for complex debugging, architectural design, and multi-file code refactoring.