OpenAI Unveils GPT‑5.4, a Spreadsheet‑Savvy Model Paving the Way for Autonomous Agents

OpenAI announced that GPT‑5.4 can read, write and reason over spreadsheets, a capability it says “opens the door to autonomous agents” across its suite of products. The model, which will be the default in ChatGPT’s new “Thinking” mode, also adds native computer‑use functions, allowing it to issue mouse clicks and keystrokes across multiple apps, according to Engadget. OpenAI claims the upgrade makes the system far better at navigating a desktop environment than its predecessor, GPT‑5.2.

The company describes GPT‑5.4 as its most capable “frontier” model for professional work, highlighting improvements in coding, data analysis, and presentation generation. Engadget notes that the model produced slides with more varied aesthetics and leveraged its image‑generation tools more effectively. The Verge adds that the model’s reasoning and coding upgrades are paired with “native computer use capabilities,” meaning it can operate a computer on a user’s behalf and complete tasks across different applications.

OpenAI positions GPT‑5.4 as a step toward an “agentic future” where AI‑powered agents run in the background to finish complex jobs online and within software. The Verge reports that the launch follows OpenAI’s earlier rollout of ChatGPT Agent, a feature that lets the chatbot act as a multi‑app assistant. By integrating spreadsheet manipulation directly into the model, OpenAI aims to reduce the need for external plugins and make autonomous workflows more reliable.

The new model also promises a clearer planning interface. When users engage GPT‑5.4 in Thinking mode, the system first outlines how it will tackle a request, giving users a chance to adjust the plan before execution, Engadget says. This “plan‑then‑act” flow is intended to improve transparency and control, especially for high‑stakes data‑analysis tasks.

OpenAI’s push comes amid intensifying competition from Google’s Gemini 3 and Microsoft’s agent modes for Excel and Word, which Ars Technica describes as “vibe‑style” coding tools for knowledge work. Reuters previously reported OpenAI’s rapid development cycle after a “code red” memo to counter Google’s advances. GPT‑5.4, with its spreadsheet fluency and native computer‑use, marks the latest escalation in that race.