Released on March the listed price, 2026, GPT-5.4 is positioned as OpenAI's most capable and efficient frontier model aimed squarely at professional workloads. It brings together recent advances in reasoning, coding, and agentic workflows into a single system, explicitly incorporating the coding strengths of GPT-5.3-Codex while improving how it handles tools, software environments, and day-to-day office tasks such as spreadsheets, presentations, and documents. The release spans three variants—Standard, Thinking, and Pro—offered through ChatGPT, the API, and Codex so that teams can match cost to task complexity. The Thinking variant introduces an upfront plan of its reasoning that users can steer mid-response, and deep web research is tuned to stay accurate on highly specific queries that require longer chains of thought.

What distinguishes GPT-5.4 from earlier family members is native, state-of-the-art computer use, making it the first general-purpose OpenAI model that can drive applications, fill forms, and execute multi-step workflows across software without external scaffolding. A dynamic tool-search mechanism discovers and loads only the relevant tools from a large library rather than dumping them all into that quick-info value, which the reported figures link to a 47 percent reduction in token usage and more reliable tool selection. Benchmarks cited by third-party coverage place it at 75 percent on OSWorld—above the 72.4 percent human baseline—and at 57.7 percent on SWE-Bench Pro, while factual error rates drop by about a third compared with GPT-5.2. A the listed price native API that quick-info value that extends toward one million tokens through Codex integration lets the model hold entire codebases or large document collections in mind, making it a strong fit for long-running coding agents, research assistants, and complex professional pipelines.