OpenAI Unveils GPT‑5.4 with Native Computer Use, Powering ChatGPT for Excel’s

OpenAI rolled out GPT‑5.4 this week, branding it the first “general‑purpose” model with native computer‑use capabilities. The company says the new mode eliminates the need for external tool‑calling APIs that previous models relied on, allowing the model to interact directly with operating‑system functions such as file I/O, spreadsheet manipulation and web requests (OpenAI report).

The debut of GPT‑5.4 is tied to a beta add‑in for Microsoft Excel called “ChatGPT for Excel.” The add‑in lets users generate formulas, edit cells and run scenario analyses by typing natural‑language prompts. OpenAI claims the model is “specifically optimized for financial tasks like modeling, scenario analysis, and data evaluation” (Matthias Bastian, The‑Decoder). Early testing on an internal benchmark of investment‑banking workflows shows GPT‑5.4 “Thinking” scoring 0.873, outpacing its predecessor GPT‑5.2 Pro (0.717) and even the competing Opus 4.6 model (0.641) (The‑Decoder).

In addition to the Excel integration, OpenAI announced financial‑data connectors for FactSet, Moody’s, S&P Global and LSEG, enabling the model to pull live market figures into its calculations. The feature is currently limited to Business, Enterprise, Pro and Plus subscribers in the United States, Canada and Australia, with a Google‑Sheets version slated for later release (The‑Decoder).

Industry observers note that native computer use is a critical step toward fully autonomous AI agents. The Verge highlighted GPT‑5.4 as “a big step toward autonomous agents,” suggesting that future iterations could execute end‑to‑end workflows without human intervention (The Verge). VentureBeat echoed this view, emphasizing the model’s “financial plugins” and its potential to streamline complex spreadsheet tasks for corporate users (VentureBeat).

Technical details released by OpenAI indicate that GPT‑5.4 also expands the context window to one million tokens and revises the tool‑calling architecture to be internally managed rather than exposed as separate API calls (The Next Web). The company has not published code samples yet, but the documentation states that developers can invoke the native computer‑use mode through a single parameter flag, simplifying integration for third‑party applications.

Analysts will watch how quickly enterprises adopt the new capabilities, especially given the model’s strong performance on professional benchmarks. If the native computer‑use claim holds up in real‑world deployments, GPT‑5.4 could set a new baseline for AI‑driven productivity tools across finance, analytics and beyond.