AI News (2026/3/6): OpenAI's Most Advanced Model GPT-5.4 Officially Launched: Native Support for Computer Control
Executive Summary:
OpenAI has officially launched GPT-5.4, which for the first time integrates reasoning, programming, native computer interaction, deep web search, and million-token context handling capabilities in a single model. The model scored 83% in the GDPval knowledge work test, 57.7% in the SWE-Bench Pro programming test, and 75% in the OSWorld computer operation test, all of which are industry-leading.
OpenAI's Most Advanced Model GPT-5.4 Officially Launched: Native Support for Computer Control
News Details
OpenAI officially released its latest model, GPT-5.4, on Friday, March 6. This model integrates reasoning, programming, native computer interaction, deep web search, and million-token context handling capabilities within a single architecture. The combination of these features has enabled GPT-5.4 to perform exceptionally well in multiple benchmark tests, particularly in the GDPval knowledge work test, SWE-Bench Pro programming test, and OSWorld computer operation test, achieving scores of 83%, 57.7%, and 75%, respectively.
Key Points
Reasoning Ability: GPT-5.4 scored 83% in the GDPval knowledge work test, demonstrating its strong capability in handling complex reasoning tasks. This score indicates that the model can better understand and process multi-step logical problems.
Programming Ability: In the SWE-Bench Pro programming test, GPT-5.4 achieved an accuracy rate of 57.7%. This improvement allows developers to more efficiently generate and debug code, reducing the time and errors associated with manual coding.
Native Computer Interaction: GPT-5.4 scored 75% in the OSWorld computer operation test, indicating that it can directly interact with operating systems to perform file operations, system commands, and other tasks. Additionally, the model supports deep web search and million-token context handling, further enhancing its flexibility and practicality in real-world applications.
AI-ALL In-Depth Review
The release of GPT-5.4 marks another significant advancement by OpenAI in the field of multimodal AI models. By integrating reasoning, programming, native computer interaction, deep web search, and million-token context handling capabilities into a single model, OpenAI has not only improved the overall performance of the model but also provided developers and businesses with a more powerful tool. The introduction of this model will have a profound impact on the existing AI ecosystem, especially in the areas of automated tasks and knowledge work. The high scores achieved by GPT-5.4 not only reflect technological progress but also open up more possibilities for future AI applications.