On March 5, OpenAI introduced its latest artificial intelligence model, GPT-5.4, which is being touted as the most powerful and efficient version to date.
In addition to the standard model, users can access two variants: GPT-5.4 Thinking, designed for cognitive tasks, and GPT-5.4 Pro, optimized for high-performance applications.
The developer edition can process up to one million tokens simultaneously, marking it as the highest data capacity among OpenAI’s offerings.
Moreover, GPT-5.4 has demonstrated significant improvements in resource efficiency, performing the same tasks while consuming considerably less computational power compared to its predecessor.
The model has achieved record results in tests assessing its ability to handle computer-related tasks, as well as general knowledge and professional competencies.
According to Brendan Foody, CEO of Mercor, GPT-5.4 has excelled in professional knowledge assessments, particularly in the fields of law and finance. The model has also become more accurate, with a 33% reduction in factual errors and an 18% decrease in general inaccuracies compared to the previous version.
OpenAI has also enhanced the technical framework of the model by introducing a Tool Search feature. This allows the AI to load instructions for tools only as needed, thereby increasing the system’s speed and reducing operational costs.
Safety features have been prioritized, with developers implementing a “thought process” verification system. This allows for visibility into the AI’s internal reasoning during complex tasks, ensuring that the neural network does not obscure its logic from developers.
“Deception is less likely in the Thinking version of GPT-5.4, indicating that the model lacks the ability to conceal its reasoning, and monitoring the thought chain remains an effective safety tool,” stated OpenAI.
OpenAI's GPT-5.4 model introduces significant advancements in processing capacity and efficiency, alongside enhanced accuracy and safety features. The new model aims to improve performance in professional fields while reducing operational costs through innovative technical updates.