Wedoany.com Report on Mar 6th, Artificial intelligence company OpenAI in the United States has released the latest AI model, GPT-5.4, just two days after the launch of its predecessor, GPT-5.3 Instant. The GPT-5.4 AI model offers two versions: Thinking and Pro, both available through paid API and the Codex application. The Thinking version is aimed at ChatGPT Plus subscribers, while the Pro version is specifically designed for enterprise users. Free users may also experience this model during specific queries.
The main highlights of the GPT-5.4 AI model include efficiency improvements, with OpenAI reporting a 47% reduction in token usage for certain tasks. The model supports a native computer use mode, enabling it to operate computers and cross-application workflows like a human. Additionally, the newly introduced ChatGPT integration allows GPT-5.4 to be directly embedded into Microsoft Excel and Google Sheets, facilitating financial analysis and automation.
OpenAI stated that GPT-5.4 supports contexts of up to 1 million tokens in the API, but costs double when inputs exceed 272,000 tokens. Benchmark tests show that in the BrowseComp evaluation, GPT-5.4 improved by 17% compared to its predecessor, with the Pro version achieving an 89.3% success rate. In the OSWorld-Verified test, GPT-5.4 achieved a 75.0% success rate, close to human performance at 72.4%.
The model also introduces a tool search function, reducing token usage and improving efficiency. In the financial sector, new products driven by GPT-5.4 include ChatGPT for Excel and Google Sheets beta versions, aimed at unifying market data and workflows. Early tester Daniel Swiecki said, "In internal financial evaluations, GPT-5.4 improved accuracy by 30 percentage points." Mercor CEO Brendan Foody called GPT-5.4 the best model the company has ever tried.
The GPT-5.4 AI model is positioned to go beyond simple Q&A, focusing on continuous professional workflows, including tool orchestration and long-context processing. OpenAI emphasized its improved factual accuracy, with the likelihood of individual statement errors being 33% lower than the previous generation. Despite higher pricing, a spokesperson stated that its enhanced capabilities and efficiency improvements keep it competitive in the market.
