The U.S. artificial intelligence (AI) company said the model integrates recent advancements in reasoning, coding, and workflows, building on the capabilities of GPT-5.3-Codex.
OpenAI also introduced GPT-5.4 Pro, which it described as a more powerful version for complex tasks.
GPT-5.4 improves performance in spreadsheets, presentations, and documents, allowing for more efficient completion of professional work, said the company, adding the model also supports larger context processing and better web search integration.
The GPT-5.4 Thinking version in ChatGPT can outline its reasoning before providing a final answer, allowing users to adjust instructions during the response process. It also recognizes interface elements from screenshots, enabling it to perform mouse and keyboard operations across different software.
In benchmark tests, GPT-5.4 outperformed previous models. It matched or exceeded industry professionals in 83 percent of GDPval comparisons and achieved higher scores than its predecessor in coding ability tests, it said.
The model has improved factual accuracy and is less likely to produce incorrect information compared to GPT-5.2.
GPT-5.4 is available to ChatGPT Plus, Team, and Pro users, and to developers through the API. OpenAI plans to gradually retire GPT-5.2 Thinking in the coming months. Despite higher per-token costs, improved token efficiency may lower overall costs in some scenarios, it said.
