GPT-5.4 mini (xhigh) is a high-efficiency reasoning model released by OpenAI on March 17, 2026. Positioned as a faster and more cost-effective alternative to the flagship GPT-5.4, it is designed to deliver frontier-level intelligence for high-volume workloads and agentic workflows. The '(xhigh)' designation refers to the model's highest reasoning effort setting, enabling it to perform deep, multi-step problem solving that approaches the capabilities of larger models while maintaining significantly lower latency.
The model features substantial improvements over the previous GPT-5 mini, running more than twice as fast while narrowing the performance gap with the full GPT-5.4 model. It excels in complex tasks such as coding, professional reasoning, and multimodal understanding. Benchmark results indicate that the model performs competitively on industry standards like SWE-Bench Pro and OSWorld-Verified, making it suitable for production environments that require a balance of high-level logic and operational efficiency.
Equipped with advanced multimodal capabilities, GPT-5.4 mini (xhigh) can process both text and image inputs. It is specifically optimized for 'computer use' tasks, allowing it to interpret dense user interface screenshots and navigate software environments autonomously. Its architecture is further refined for tool use and function calling, which supports the orchestration of complex subagents that handle specialized execution tasks under the guidance of a primary planning model.
In addition to its reasoning performance, the model supports a massive 400,000-token context window, allowing for the ingestion and analysis of extensive codebases or document sets in a single pass. It is utilized across OpenAI’s product suite as a primary reasoning engine for developers and a fallback for the flagship models during peak usage, ensuring consistent access to advanced thinking capabilities.