OpenAI Launches GPT-5.4: Native Computer Operation Capabilities, Surpassing Human Performance in Professional Scenarios

OpenAI recently launched GPT-5.4, an AI tool defined as the 'cutting-edge model for professional work' with breakthrough capabilities like native computer operation, 1M context window, and tool search. It can directly operate computer interfaces, complete workflows such as sending emails and creating schedules, clicking mice and typing keyboards like a real person. In performance tests, GPT-5.4 showed outstanding results: 75% success rate in OSWorld desktop navigation, surpassing humans' 72.4%; 87.3% score in Excel modeling, nearly 20 percentage points higher than the previous generation; 68% of human reviewers preferred its PPTs. Additionally, its hallucination rate decreased, with false statements reduced by 33%. Although the API unit price is higher than the previous generation, the tool search function reduces Token consumption by 47%, making the overall cost more optimal. Currently, GPT-5.4 has gradually been made available to ChatGPT Plus, Team, and Pro users, with Pro users also able to use the more powerful GPT-5.4 Pro version.