OpenAI releases GPT-5.4 model with advanced reasoning, coding, and native computer use

OpenAI has launched its latest model, GPT-5.4, which integrates cutting-edge reasoning skills, enhanced coding functionalities, and superior task performance into a single flagship product. This new release includes two specialized variants: GPT-5.4 Thinking, designed for advanced reasoning, and GPT-5.4 Pro, optimized for tackling complex tasks with maximum efficiency. According to a blog post by OpenAI, the new model retains the coding capabilities of its predecessor, GPT-5.3-Codex, while significantly enhancing its performance across various tools and software applications, including spreadsheets, presentations, and document processing. One of the standout features of GPT-5.4 Thinking in ChatGPT is its ability to showcase its reasoning process before finalizing a response. This transparency allows users to make adjustments while the model is generating answers, resulting in more accurate and customized outputs. Additionally, the API version boasts an impressive context window of 1 million tokens, greatly expanding its operational capacity. OpenAI has emphasized that the new model incorporates “computer-use capabilities,” enabling it to operate computers and execute intricate workflows across multiple applications. Compared to its predecessor, GPT-5.2, GPT-5.4 demonstrates a remarkable reduction in token usage to reach similar conclusions, establishing itself as the most efficient reasoning model to date in terms of both speed and cost. In terms of accuracy, GPT-5.4 is reported to be the most factual model available, with a 33% lower likelihood of false statements and 18% fewer errors in full responses compared to GPT-5.2. The model achieved a notable score of 83% on OpenAI's GDPval test, a significant improvement over the 70.9% score of GPT-5.2, which assesses the model’s capacity to generate precise knowledge work across 44 different occupations. Further demonstrating its capabilities, GPT-5.4 scored 87.3% on internal benchmark tests related to spreadsheet modeling tasks, outperforming the previous version by a considerable margin. Brendan Foody, CEO of Mercor, highlighted that GPT-5.4 excels in producing extensive deliverables such as slide presentations, financial models, and legal analyses, all while operating faster and more cost-effectively than competing models. In coding, GPT-5.4 reportedly achieves up to 1.5 times faster token velocity, streamlining the processes of coding, iteration, and debugging. For users of ChatGPT, GPT-5.4 Thinking is accessible to Plus, Team, and Pro subscribers, while the legacy GPT-5.2 Thinking will continue to be available for another three months before its retirement on June 5, 2026. The Pro version is specifically catered to Pro and Enterprise users. API pricing for GPT-5.4 is set at $2.50 per million input tokens and $15 per million output tokens, whereas GPT-5.4 Pro is priced at $30 per million input tokens and $180 per million output tokens.